The wire
Material changes only, each carrying the date it was observed and the condition the figure holds under. Frontier movement, price moves at or over 5% on the balanced workload blend, advertised context changes, and published corrections. A new model listing is not a wire event, and neither is anything we merely have an opinion about.
26 material changes
In the 7 days to 2026-11-05 (from 2026-10-30), counted from the events below rather than stamped on the page.
What is not here, and why
- No frontier movement is published for 2026-08-25 → 2026-08-26. The ranked board changed by 62% between those two snapshots — 56 slugs entered the ranked set and 44 left, of a 162-slug union — which is over the 20% ceiling this wire treats as a definition change rather than a market move. The frontier is a Pareto set over that board, so it moves when the board does. Price movement is unaffected: a per-model price delta does not depend on who else is ranked.
- 2 models moved between having a balanced blend price and not having one between 2026-09-30 and 2026-10-06, or moved off a zero baseline. They are not in the feed: there is no percentage to compute, and treating an absent price as zero is how a model that lost its price reads as one that became free.
- Changes dated 2026-10-06 are measured against the 2026-09-30 day file, which is kept as that morning's release published it (03:34 UTC). A later release the same day (16:45 UTC) had already shown some of them: 9 of the 12 frontier moves dated 2026-10-06 were on the site from that release. One change since that release has no event here: Gemma 4 26B A4B is on the frontier again.
- 1 model moved between having a balanced blend price and not having one between 2026-10-13 and 2026-10-14, or moved off a zero baseline. It is not in the feed: there is no percentage to compute, and treating an absent price as zero is how a model that lost its price reads as one that became free.
- DeepSeek: DeepSeek V3's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V3 0324's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V3.1's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V3.2's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V4 Flash 0423's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V4 Flash Vision Exp's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V4 Pro 0813's price move on 2026-10-06 is not published: it is this site's restatement of the headline to the standard price window (R106, §83 of docs/EMPIRICAL-GROUNDING.md), not a price the vendor changed.
- DeepSeek: DeepSeek V4 Pro 0813's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- DeepSeek: DeepSeek V4.1 Flash's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Google: Gemini 3.7 Flash's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Google: Gemini 3.8 Flash's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Google: Gemma 3 27B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Google: Gemma 4 31B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Inception: Mercury 2.5's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- inclusionAI: Ling 3.0 Flash's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- inclusionai/ling-3.0-flash-fin's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- inclusionAI: Ling 3.0 Flash VL's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Meituan: LongCat 2.0's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Meta: Llama 3.1 70B Instruct's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Meta: Llama 3.1 8B Instruct's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Meta: Llama 3.3 70B Instruct's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Meta: Llama 4 Maverick's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MiniMax: MiniMax M2.5's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MiniMax: MiniMax M2.7's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MiniMax: MiniMax M3's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Mistral: Mistral Small 3.2 24B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MoonshotAI: Kimi K2.5's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MoonshotAI: Kimi K2.6's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MoonshotAI: Kimi K3's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- NVIDIA: Nemotron 3 Super's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- NVIDIA: Nemotron 3 Ultra's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- OpenAI: GPT-5.6 Sol's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- OpenAI: GPT-5.6 Sol Pro's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- OpenAI: gpt-oss-120b's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- poolside/laguna-s-2.1's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- poolside/laguna-xs-2.1's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3 30B A3B Instruct 2507's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3 Coder 480B A35B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3 Next 80B A3B Instruct's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3 VL 235B A22B Instruct's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3 VL 30B A3B Instruct's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3 VL 30B A3B Thinking's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3.5-35B-A3B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3.5 397B A17B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3.6 35B A3B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Qwen: Qwen3.8 27B's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Tencent: Hy3's price move on 2026-10-06 is not published: it is this site's restatement of the headline to the standard price window (R106, §83 of docs/EMPIRICAL-GROUNDING.md), not a price the vendor changed.
- Upstage: Solar Mini 4's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Upstage: Solar Pro 4's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Z.ai: GLM 4.7's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Z.ai: GLM 5's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Z.ai: GLM 5.1's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- Z.ai: GLM 5.3's price move on 2026-10-06 is not published: this site changed what its price is, from OpenRouter's model-level price to the price the reference-price rule takes from its endpoint rows (§96 of docs/EMPIRICAL-GROUNDING.md). A seller's move in the same days is inside it.
- MythoMax 13B's price move on 2026-10-14 is not published: this site changed which offer sets its price, by adding a context test to the reference-price rule (§100 of docs/EMPIRICAL-GROUNDING.md). If a seller changed its price in the same days, that move is inside it.
- Meta: Llama 3.3 70B Instruct's price move on 2026-10-14 is not published: this site changed which offer sets its price, by adding a context test to the reference-price rule (§100 of docs/EMPIRICAL-GROUNDING.md). If a seller changed its price in the same days, that move is inside it.
- Meta: Llama 4 Scout's price move on 2026-10-14 is not published: this site changed which offer sets its price, by adding a context test to the reference-price rule (§100 of docs/EMPIRICAL-GROUNDING.md). If a seller changed its price in the same days, that move is inside it.
- Qwen: Qwen3 32B's price move on 2026-10-14 is not published: this site changed which offer sets its price, by adding a context test to the reference-price rule (§100 of docs/EMPIRICAL-GROUNDING.md). If a seller changed its price in the same days, that move is inside it.
- Qwen: Qwen3.8 27B's price move on 2026-10-15 is not published: this site changed which offer sets its price, by adding a context test to the reference-price rule (§100 of docs/EMPIRICAL-GROUNDING.md). If a seller changed its price in the same days, that move is inside it.
- 1 price movement is not published for DeepSeek: DeepSeek V4.1 Flash 1017 after 2026-11-02. Its price is held at the live record of that date, because the upstream figure since was reviewed and not accepted (data/upstream-holds.json). A snapshot written before the hold records that figure; it is not a price this site showed.
- Context events are advertised-window changes. Context-tier ladders — the price cliffs — are not in the wire: neither the rank snapshots nor data/upstream-baseline.json keeps tier history, so a cliff introduced today cannot be told apart from one we never recorded.
- No context event is published for Xiaomi: MiMo-V2.5 yet. The feed shows another context window than the one this site publishes, and the published one is kept until a later day shows the same change (data/upstream-holds.json). The event is published the day the site changes its figure.
- 16 corrections in docs/EMPIRICAL-GROUNDING.md carry no date stamp and are therefore not in the wire. An entry with no date cannot be placed on a timeline. The log itself is complete; this feed is the incomplete view of it.
- Earlier corrections from the correction log that the wire does not carry: 51. Correction log
2026-11-05
price
Xiaomi: MiMo-V2.6-Flash is 20.7% dearer on the balanced blend
$0.145/M on 2026-11-04, $0.175/M on 2026-11-05 — a 20.7% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: xiaomi/mimo-v2.6-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-11-04 → 2026-11-05. Threshold: 5% or more, the same one watch-upstream routes to a human.
context
Qwen: Qwen3.8 27B advertised context changed from 1000000 to 1024000 tokens
Advertised context window 1000000 → 1024000 tokens on 2026-11-05. The window moved; the price column does not record that it did. Figures are printed in the units the feed publishes them in, not abbreviated. Model: qwen/qwen3.8-27b.
Advertised context window from the upstream models feed, diffed against data/upstream-baseline.json and checked 2026-11-05T07:06:25.875Z. Standard delivery only.
2026-11-04
price
DeepSeek: DeepSeek V4.1 Flash is 48.8% dearer on the balanced blend
$0.0645/M on 2026-11-02, $0.096/M on 2026-11-04 — a 48.8% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: deepseek/deepseek-v4.1-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-11-02 → 2026-11-04. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Xiaomi: MiMo-V2.5-Pro is 49% dearer on the balanced blend
$0.5437/M on 2026-11-02, $0.81/M on 2026-11-04 — a 49% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: xiaomi/mimo-v2.5-pro.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-11-02 → 2026-11-04. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Z.ai: GLM 5.3 Flash is 42.4% cheaper on the balanced blend
$0.1775/M on 2026-11-02, $0.1023/M on 2026-11-04 — a 42.4% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: z-ai/glm-5.3-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-11-02 → 2026-11-04. Threshold: 5% or more, the same one watch-upstream routes to a human.
correction
Correction: A test counted the archives a refresh had already written, and stopped the first daily run after the rankings release
Logged 2026-11-04 as §132 of docs/EMPIRICAL-GROUNDING.md. The wire carries the fact that a correction was published; the log carries what was claimed, what is true, and how it was found. Corrections are not summarised here — shortening one is how a load-bearing qualification gets dropped. A vendor changing a price is not us being wrong; this is.
Correction log, docs/EMPIRICAL-GROUNDING.md §132, logged 2026-11-04.
2026-11-01
price
MiniMax: MiniMax M3.1 Flash is 66.7% cheaper on the balanced blend
$0.2625/M on 2026-10-31, $0.0875/M on 2026-11-01 — a 66.7% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: minimax/minimax-m3.1-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-31 → 2026-11-01. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
MoonshotAI: Kimi K3 is 12.4% cheaper on the balanced blend
$4.52/M on 2026-10-31, $3.96/M on 2026-11-01 — a 12.4% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: moonshotai/kimi-k3.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-31 → 2026-11-01. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Particle AI: Particle-1 is 62.9% cheaper on the balanced blend
$0.4375/M on 2026-10-31, $0.1625/M on 2026-11-01 — a 62.9% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: particle-ai/particle-1.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-31 → 2026-11-01. Threshold: 5% or more, the same one watch-upstream routes to a human.
2026-10-31
frontier join
Z.ai: GLM 5.3 joined the frontier
1473.8 on LMArena at $0.4145/M on 2026-10-31, against 1473.8 at $0.875/M on 2026-10-30. Model: z-ai/glm-5.3.
Default board: standard delivery, not retired, LMArena general axis, balanced-workload blend. Rank snapshots 2026-10-30 → 2026-10-31.
frontier leave
DeepSeek: DeepSeek V4 Flash 0423 left the frontier
1431.8 on LMArena at $0.1125/M on 2026-10-31, unchanged from 2026-10-30. Its own figures did not move — the board did. A frontier is a Pareto set, so a model leaves it when the rest of the board changes around it. The wire does not name what now beats it: the dominance rule refuses a replacement that costs capability, and the snapshot projection carries no context window or modalities to check that with. Model: deepseek/deepseek-v4-flash.
Default board: standard delivery, not retired, LMArena general axis, balanced-workload blend. Rank snapshots 2026-10-30 → 2026-10-31.
frontier leave
Google: Gemma 4 26B A4B left the frontier
1434.2 on LMArena at $0.1275/M on 2026-10-31, unchanged from 2026-10-30. Its own figures did not move — the board did. A frontier is a Pareto set, so a model leaves it when the rest of the board changes around it. The wire does not name what now beats it: the dominance rule refuses a replacement that costs capability, and the snapshot projection carries no context window or modalities to check that with. Model: google/gemma-4-26b-a4b-it.
Default board: standard delivery, not retired, LMArena general axis, balanced-workload blend. Rank snapshots 2026-10-30 → 2026-10-31.
frontier leave
Xiaomi: MiMo-V2.6-Flash left the frontier
1451.2 on LMArena at $0.145/M on 2026-10-31, unchanged from 2026-10-30. Its own figures did not move — the board did. A frontier is a Pareto set, so a model leaves it when the rest of the board changes around it. The wire does not name what now beats it: the dominance rule refuses a replacement that costs capability, and the snapshot projection carries no context window or modalities to check that with. Model: xiaomi/mimo-v2.6-flash.
Default board: standard delivery, not retired, LMArena general axis, balanced-workload blend. Rank snapshots 2026-10-30 → 2026-10-31.
price
DeepSeek: DeepSeek V4.1 Flash is 60.4% cheaper on the balanced blend
$0.1628/M on 2026-10-30, $0.0645/M on 2026-10-31 — a 60.4% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: deepseek/deepseek-v4.1-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-30 → 2026-10-31. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Mistral: Mistral Large 4 is 49.2% cheaper on the balanced blend
$2.07/M on 2026-10-30, $1.05/M on 2026-10-31 — a 49.2% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: mistralai/mistral-large-4-0.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-30 → 2026-10-31. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
NVIDIA: Nemotron 3.5 Super VL is 61.1% cheaper on the balanced blend
$0.45/M on 2026-10-30, $0.175/M on 2026-10-31 — a 61.1% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: nvidia/nemotron-3.5-super-vl.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-30 → 2026-10-31. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Qwen: Qwen3.5-27B is 27.8% cheaper on the balanced blend
$0.7425/M on 2026-10-30, $0.5363/M on 2026-10-31 — a 27.8% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: qwen/qwen3.5-27b.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-30 → 2026-10-31. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Z.ai: GLM 5.2 is 72.1% cheaper on the balanced blend
$1.48/M on 2026-10-30, $0.4145/M on 2026-10-31 — a 72.1% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: z-ai/glm-5.2.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-30 → 2026-10-31. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Z.ai: GLM 5.3 is 52.6% cheaper on the balanced blend
$0.875/M on 2026-10-30, $0.4145/M on 2026-10-31 — a 52.6% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: z-ai/glm-5.3.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-30 → 2026-10-31. Threshold: 5% or more, the same one watch-upstream routes to a human.
correction
Correction: A board LMArena had not republished read as 48 days stale on the first day its read failed
Logged 2026-10-31 as §131 of docs/EMPIRICAL-GROUNDING.md. The wire carries the fact that a correction was published; the log carries what was claimed, what is true, and how it was found. Corrections are not summarised here — shortening one is how a load-bearing qualification gets dropped. A vendor changing a price is not us being wrong; this is.
Correction log, docs/EMPIRICAL-GROUNDING.md §131, logged 2026-10-31.
2026-10-30
price
DeepSeek: DeepSeek V4.1 Flash 1017 is 57.1% cheaper on the balanced blend
$0.3/M on 2026-10-29, $0.1288/M on 2026-10-30 — a 57.1% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: deepseek/deepseek-v4.1-flash-1017.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-29 → 2026-10-30. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
DeepSeek: DeepSeek V4 Pro 0423 is 5.6% cheaper on the balanced blend
$1.63/M on 2026-10-29, $1.53/M on 2026-10-30 — a 5.6% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: deepseek/deepseek-v4-pro.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-29 → 2026-10-30. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
NVIDIA: Nemotron 3 Super is 403.7% dearer on the balanced blend
$0.1638/M on 2026-10-29, $0.825/M on 2026-10-30 — a 403.7% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: nvidia/nemotron-3-super-120b-a12b.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-29 → 2026-10-30. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
NVIDIA: Nemotron 3 Ultra is 35.1% dearer on the balanced blend
$0.925/M on 2026-10-29, $1.25/M on 2026-10-30 — a 35.1% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: nvidia/nemotron-3-ultra-550b-a55b.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-29 → 2026-10-30. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Z.ai: GLM 5.3 Flash is 25.3% cheaper on the balanced blend
$0.2375/M on 2026-10-29, $0.1775/M on 2026-10-30 — a 25.3% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: z-ai/glm-5.3-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-29 → 2026-10-30. Threshold: 5% or more, the same one watch-upstream routes to a human.
context
NVIDIA: Nemotron 3 Super advertised context changed from 262144 to 1048576 tokens
Advertised context window 262144 → 1048576 tokens on 2026-10-30. The window moved; the price column does not record that it did. Figures are printed in the units the feed publishes them in, not abbreviated. Model: nvidia/nemotron-3-super-120b-a12b.
Advertised context window from the upstream models feed, diffed against data/upstream-baseline.json and checked 2026-10-30T07:05:38.737Z. Standard delivery only.
2026-10-29
price
MoonshotAI: Kimi K3 is 48.3% dearer on the balanced blend
$3.05/M on 2026-10-28, $4.52/M on 2026-10-29 — a 48.3% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: moonshotai/kimi-k3.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-28 → 2026-10-29. Threshold: 5% or more, the same one watch-upstream routes to a human.
2026-10-28
frontier join
Xiaomi: MiMo-V2.6-Flash joined the frontier
1451.7 on LMArena at $0.145/M on 2026-10-28, against 1451.7 at $0.1675/M on 2026-10-27. Model: xiaomi/mimo-v2.6-flash.
Default board: standard delivery, not retired, LMArena general axis, balanced-workload blend. Rank snapshots 2026-10-27 → 2026-10-28.
price
DeepSeek: DeepSeek V4 Pro 0423 is 34.6% dearer on the balanced blend
$1.21/M on 2026-10-27, $1.63/M on 2026-10-28 — a 34.6% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: deepseek/deepseek-v4-pro.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
IBM: Granite 4.2 8B is 23.3% cheaper on the balanced blend
$0.1075/M on 2026-10-27, $0.0825/M on 2026-10-28 — a 23.3% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: ibm-granite/granite-4.2-8b.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
MoonshotAI: Kimi K2.6 is 69.7% dearer on the balanced blend
$0.825/M on 2026-10-27, $1.40/M on 2026-10-28 — a 69.7% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: moonshotai/kimi-k2.6.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
MoonshotAI: Kimi K2.7 Code is 18.9% dearer on the balanced blend
$1.34/M on 2026-10-27, $1.59/M on 2026-10-28 — a 18.9% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: moonshotai/kimi-k2.7-code.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
NVIDIA: Nemotron 3.5 Super VL is 89.5% dearer on the balanced blend
$0.2375/M on 2026-10-27, $0.45/M on 2026-10-28 — a 89.5% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: nvidia/nemotron-3.5-super-vl.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
NVIDIA: Nemotron 3 Ultra is 26% cheaper on the balanced blend
$1.25/M on 2026-10-27, $0.925/M on 2026-10-28 — a 26% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: nvidia/nemotron-3-ultra-550b-a55b.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Reflection: Beam 1.0 501B A23B is 37.1% dearer on the balanced blend
$0.4375/M on 2026-10-27, $0.6/M on 2026-10-28 — a 37.1% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: reflection/beam-1.0-501b-a23b.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Xiaomi: MiMo-V2.5 is 357.1% dearer on the balanced blend
$0.175/M on 2026-10-27, $0.8/M on 2026-10-28 — a 357.1% rise in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: xiaomi/mimo-v2.5.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
price
Xiaomi: MiMo-V2.6-Flash is 13.4% cheaper on the balanced blend
$0.1675/M on 2026-10-27, $0.145/M on 2026-10-28 — a 13.4% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: xiaomi/mimo-v2.6-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-27 → 2026-10-28. Threshold: 5% or more, the same one watch-upstream routes to a human.
context
inclusionAI: Ling 3.1 Flash advertised context changed from 262144 to 1000000 tokens
Advertised context window 262144 → 1000000 tokens on 2026-10-28. The window moved; the price column does not record that it did. Figures are printed in the units the feed publishes them in, not abbreviated. Model: inclusionai/ling-3.1-flash.
Advertised context window from the upstream models feed, diffed against data/upstream-baseline.json and checked 2026-10-28T07:05:26.754Z. Standard delivery only.
2026-10-27
frontier leave
Xiaomi: MiMo-V2.6-Flash left the frontier
1451.7 on LMArena at $0.1675/M on 2026-10-27, against 1455.2 at $0.1675/M on 2026-10-21. Model: xiaomi/mimo-v2.6-flash.
Default board: standard delivery, not retired, LMArena general axis, balanced-workload blend. Rank snapshots 2026-10-21 → 2026-10-27.
price
DeepSeek: DeepSeek V4.1 Flash is 10.6% cheaper on the balanced blend
$0.1822/M on 2026-10-21, $0.1628/M on 2026-10-27 — a 10.6% fall in the blended figure. The blend weights input and output, so the per-million prices behind it may have moved by different amounts and one of them may not have moved at all. This is the site's published price: the model's reference price where one of its offers passes the like-for-like test, and otherwise a labelled fallback (see /methodology/). It moves when the offer it is taken from changes its rate or another offer takes its place. It is not necessarily the model vendor's price. Model: deepseek/deepseek-v4.1-flash.
Balanced-workload blend, effective price per million tokens, standard delivery. Rank snapshots 2026-10-21 → 2026-10-27. Threshold: 5% or more, the same one watch-upstream routes to a human.
Showing the newest 40 of 200. The rest are in the feed and the JSON.
Where this comes from
Events are computed from the dated rank snapshots in data/snapshots/, a
diff of the upstream models feed, and the correction log. Frontier movement is only
published when the ranked board itself held still between two snapshots — a frontier is
a Pareto set, so a board that changes underneath it moves the answer on its own, and
that is a definition change rather than a market one.
Scores cited are LMArena Elo, used under CC BY 4.0. No Artificial Analysis figures appear on this page or in the feed: their redistribution terms are unresolved. Unrated is not zero. A model with no independent score is named as unrated, never scored.