Undominated.ai

Spreads worth shopping

Ranked by actionable input-price spread across distinct sellers at equal-or-better serving precision, where the cheaper seller is also cheaper on the balanced blend (3 parts input to 1 part output). A vendor’s own service tiers are not competition. Unpriced models are excluded. Unrated is not scored zero.

45 models

As of . Condition: equal-or-better declared precision; cheaper lower-precision is not a saving; the cheaper input seller is also cheaper on the balanced blend (3 input : 1 output).

Worth shopping

Spread is the dearest ÷ cheapest input price among sellers at the reference precision or better, listed only where the cheaper seller is also cheaper on the balanced blend. Cheaper lower-precision endpoints do not enter this table as a saving.

ModelInput spreadSellersCheapest inputDearest inputPrecision
gpt-oss-120b 11.67×5$0.030 DekaLLM$0.350 Cerebrasbf16
GLM 5.3 lower-quant also listed11.05×11$0.190 AkashML$2.10 BaseTenfp8
DeepSeek V4 Flash 0731 lower-quant also listed10×10$0.044 StreamLake$0.440 Baidufp8
DeepSeek V4 Pro 0423 lower-quant also listed8.09×8$0.209 StreamLake$1.69 Baidufp8
Qwen3.8 27B lower-quant also listed6.6×2$0.150 DeepInfra$0.990 Cerebrasbf16
DeepSeek V4.1 Flash lower-quant also listed5.66×13$0.053 Morph$0.300 Venicefp8
Llama 3.3 70B Instruct lower-quant also listed5.26×2$0.135 Novita$0.710 CoreWeavebf16
MythoMax 13B 5×2$0.080 Parasail$0.400 DeepInfrafp16
GLM 5.2 lower-quant also listed4.53×12$0.309 Morph$1.40 Z.AIfp8
DeepSeek V4 Flash 0423 lower-quant also listed4.52×9$0.042 StreamLake$0.190 Mancer 2fp8
MiMo-V2.5 3.36×4$0.119 GMICloud$0.400 Venicefp8
Ling 3.0 Flash VL 2.86×2$0.021 Novita$0.060 DeepInfrabf16
GLM 5.3 Flash lower-quant also listed2.68×11$0.084 Novita$0.225 Inceptronfp8
Qwen3.6 35B A3B lower-quant also listed2.5×7$0.100 AkashML$0.250 CoreWeavefp8
DeepSeek V3.1 lower-quant also listed2.41×3$0.270 SiliconFlow$0.650 SambaNovafp8
Qwen3 235B A22B Instruct 2507 2.29×6$0.088 GMICloud$0.200 Nebiusfp8
Gemma 4 26B A4B lower-quant also listed2.21×7$0.059 Io Net$0.130 Novitabf16
DeepSeek V4 Flash Vision Exp 2.04×3$0.216 DeepInfra$0.440 SiliconFlowfp8
DeepSeek V3.2 1.77×5$0.209 GMICloud$0.371 Alibabafp8
Qwen3 32B 1.75×3$0.080 DeepInfra$0.140 SiliconFlowfp8
Llama 3.1 Euryale 70B v2.2 1.71×2$0.850 DeepInfra$1.45 Novitafp8
Qwen3.5-35B-A3B lower-quant also listed1.71×4$0.140 DeepInfra$0.240 SiliconFlowfp8
GLM 5 1.67×7$0.600 StreamLake$1.00 Z.AIfp8
Granite 4.2 8B 1.67×2$0.060 DeepInfra$0.100 CoreWeavebf16
Qwen3 235B A22B Thinking 2507 lower-quant also listed1.5×2$0.300 Novita$0.450 Venicefp8
GLM 5.1 1.45×11$0.966 StreamLake$1.40 Venicefp8
MiniMax M2.7 1.43×4$0.210 GMICloud$0.300 Minimaxfp8
R1 0528 1.4×2$0.500 SiliconFlow$0.700 Novitafp8
DeepSeek V4 Pro 0813 lower-quant also listed1.33×9$0.990 Novita$1.32 SiliconFlowfp8
Qwen3.5 397B A17B lower-quant also listed1.33×4$0.450 DeepInfra$0.600 GMICloudfp8
Llama 4 Maverick lower-quant also listed1.3×2$0.270 Novita$0.350 Parasailfp8
Llama 3 8B Lunaris 1.25×2$0.040 Parasail$0.050 Novitabf16
MiniMax M3 lower-quant also listed1.25×8$0.240 GMICloud$0.300 Minimaxfp8
Kimi K3 lower-quant also listed1.2×2$2.50 Morph$3.00 BaseTenfp8
Nemotron 3 Nano 30B A3B 1.2×2$0.050 Crusoe$0.060 Nebiusfp8
MiniMax M2 1.18×2$0.255 Minimax$0.300 Novitafp8
Nemotron 3.5 Lightning lower-quant also listed1.17×2$0.060 DeepInfra$0.070 CoreWeavebf16
DeepSeek V3 0324 1.16×2$0.250 SiliconFlow$0.290 GMICloudfp8
Kimi K2.7 Code lower-quant also listed1.11×3$0.859 SiliconFlow$0.950 Alibabafp8
Qwen3 30B A3B Instruct 2507 lower-quant also listed1.11×2$0.090 SiliconFlow$0.100 Nebiusfp8
Kimi K2.5 1.09×2$0.450 SiliconFlow$0.490 AtlasCloudint4
Qwen3 Coder 480B A35B lower-quant also listed1.09×2$0.350 Venice$0.380 Novitafp8
Qwen3.6 27B 1.08×4$0.300 Chutes$0.325 Venicefp8
gpt-oss-20b lower-quant also listed1.03×2$0.029 DekaLLM$0.030 DeepInfrabf16
MiniMax M2.5 lower-quant also listed1.02×4$0.295 AtlasCloud$0.300 Minimaxfp8

Precision differences — not established savings

These rates differ, but the cheapest endpoint has lower or undeclared precision. The raw ratio does not establish a saving at the reference precision.

ModelRaw ×Cheap (lower / unknown quant)Like-for-like
Llama 3.1 8B Instruct11×$0.020 DeepInfra fp8$0.220 CoreWeave bf16
Gemma 4 31B8.33×$0.090 DeepInfra fp4$0.140 Crusoe bf16
Mistral Nemo8.33×$0.018 DekaLLM fp8$0.023 Io Net fp16
Ling 3.0 Flash2.86×$0.021 Novita unknown$0.060 DeepInfra bf16
Llama 4 Scout2.5×$0.100 DeepInfra fp8$0.180 Novita bf16
Qwen3 VL 235B A22B Thinking2.45×$0.400 Alibaba unknown$0.980 Novita bf16
Kimi K2.62.34×$0.465 Inceptron int4$0.700 Crusoe bf16
Qwen3 14B2.28×$0.100 NextBit int4$0.120 DeepInfra fp8
Qwen3 VL 30B A3B Instruct2.23×$0.130 Alibaba unknown$0.200 Novita bf16
Qwen3 VL 8B Instruct2.14×$0.117 Alibaba unknown$0.250 Parasail bf16
Qwen3.5-9B2.13×$0.080 Darkbloom fp4$0.100 DeepInfra bf16
GLM 4.71.75×$0.400 DeepInfra fp4$0.540 Novita fp8
MiMo-V2.6-Flash1.75×$0.100 Darkbloom fp4$0.140 GMICloud bf16
Qwen3 Next 80B A3B Instruct1.67×$0.090 DeepInfra fp8$0.150 Novita bf16
Hy31.54×$0.130 DeepInfra fp4$0.140 GMICloud bf16
Qwen3.5-122B-A10B1.54×$0.260 SiliconFlow fp8$0.400 Novita bf16
Qwen3.5-27B1.54×$0.195 Alibaba unknown$0.300 Novita bf16
Qwen3 VL 235B A22B Instruct1.5×$0.200 DeepInfra fp8$0.300 Novita bf16
Gemma 3 27B1.49×$0.080 DeepInfra fp8$0.119 Novita bf16
Qwen3 VL 30B A3B Thinking1.45×$0.200 Alibaba unknown$0.290 SiliconFlow fp8
Ling 3.0 Flash Fin1.43×$0.042 Novita unknown$0.060 DeepInfra fp4
GLM 4.61.4×$0.430 Venice fp4$0.550 Novita bf16
MiniMax M11.38×$0.400 Minimax unknown$0.550 Novita bf16
Mistral Small 3.2 24B1.33×$0.075 DeepInfra fp8$0.090 Parasail bf16
ReMM SLERP 13B1.29×$0.350 Mancer 2 fp8$0.450 NextBit bf16
Nemotron 3 Ultra1.25×$0.500 DeepInfra fp4$0.625 Venice fp8
DeepSeek V31.24×$0.257 StreamLake unknown$0.320 DeepInfra fp4
GLM 4.7 Flash1.17×$0.060 Venice fp8$0.070 Novita bf16
Nemotron 3 Super1.06×$0.080 DekaLLM fp8$0.085 DeepInfra bf16
Qwen2.5 72B Instruct1.06×$0.360 DeepInfra fp8$0.380 Novita bf16
MiMo-V2.6-Pro1.01×$0.430 DeepInfra fp8$0.435 GMICloud bf16

Serving precision not established

49 models have multiple seller owners, but no declared serving precision. These price observations are excluded from the declared-precision ranking. Verify precision, context, tools, reliability and terms with the endpoint provider.

  • Claude Fable 5 · 5 sellers · $10.00–$10.00 per million input tokens
  • Claude Fable 5.1 · 4 sellers · $10.00–$10.00 per million input tokens
  • Claude Haiku 4.5 · 4 sellers · $1.00–$1.00 per million input tokens
  • Claude Opus 4.5 · 5 sellers · $5.00–$5.00 per million input tokens
  • Claude Opus 4.6 · 5 sellers · $5.00–$5.00 per million input tokens
  • Claude Opus 4.7 · 5 sellers · $5.00–$5.00 per million input tokens
  • Claude Opus 4.8 · 5 sellers · $5.00–$5.00 per million input tokens
  • Claude Opus 5 · 5 sellers · $5.00–$5.00 per million input tokens
  • Claude Opus 5.5 · 5 sellers · $4.00–$4.00 per million input tokens
  • Claude Sonnet 4.5 · 5 sellers · $3.00–$3.00 per million input tokens
  • Claude Sonnet 4.6 · 5 sellers · $3.00–$3.00 per million input tokens
  • Claude Sonnet 5 · 5 sellers · $2.00–$2.00 per million input tokens
  • Claude Sonnet 5.5 · 5 sellers · $2.00–$2.00 per million input tokens
  • GPT-3.5 Turbo 16k · 2 sellers · $3.00–$3.00 per million input tokens
  • GPT-4 · 2 sellers · $30.00–$30.00 per million input tokens
  • GPT-4.1 · 2 sellers · $2.00–$2.00 per million input tokens
  • GPT-4.1 Mini · 2 sellers · $0.400–$0.400 per million input tokens
  • GPT-4.1 Nano · 2 sellers · $0.100–$0.100 per million input tokens
  • GPT-4o · 2 sellers · $2.50–$2.50 per million input tokens
  • GPT-4o (2024-05-13) · 2 sellers · $5.00–$5.00 per million input tokens
  • GPT-4o (2024-08-06) · 2 sellers · $2.50–$2.50 per million input tokens
  • GPT-4o-mini · 2 sellers · $0.150–$0.150 per million input tokens
  • GPT-5 · 2 sellers · $1.25–$1.25 per million input tokens
  • GPT-5 Mini · 2 sellers · $0.125–$0.250 per million input tokens
  • GPT-5 Nano · 2 sellers · $0.025–$0.050 per million input tokens
  • GPT-5.1 · 2 sellers · $0.625–$1.25 per million input tokens
  • GPT-5.2 · 2 sellers · $0.875–$1.75 per million input tokens
  • GPT-5.3-Codex · 2 sellers · $1.75–$1.75 per million input tokens
  • GPT-5.4 · 3 sellers · $1.25–$2.75 per million input tokens
  • GPT-5.4 Mini · 2 sellers · $0.375–$0.750 per million input tokens
  • GPT-5.4 Nano · 2 sellers · $0.100–$0.200 per million input tokens
  • GPT-5.4 Pro · 2 sellers · $15.00–$30.00 per million input tokens
  • GPT-5.5 · 3 sellers · $2.50–$5.50 per million input tokens
  • GPT-5.6 Luna · 3 sellers · $0.100–$0.220 per million input tokens
  • GPT-5.6 Luna Pro · 2 sellers · $0.100–$0.200 per million input tokens
  • GPT-5.6 Sol · 3 sellers · $1.00–$4.40 per million input tokens
  • GPT-5.6 Sol Pro · 2 sellers · $1.00–$4.00 per million input tokens
  • GPT-5.6 Terra · 3 sellers · $1.00–$2.20 per million input tokens
  • GPT-5.6 Terra Pro · 2 sellers · $1.00–$2.00 per million input tokens
  • GPT-6 Astra · 3 sellers · $5.00–$11.00 per million input tokens
  • GPT-6 Astra Pro · 2 sellers · $5.00–$10.00 per million input tokens
  • GPT-6 Luna · 3 sellers · $0.050–$0.110 per million input tokens
  • GPT-6 Luna Pro · 2 sellers · $0.050–$0.100 per million input tokens
  • GPT-6 Sol · 3 sellers · $1.00–$2.20 per million input tokens
  • GPT-6 Sol Pro · 2 sellers · $1.00–$2.00 per million input tokens
  • GPT-6.1 Sol · 3 sellers · $1.00–$2.20 per million input tokens
  • GPT-6.1 Sol Pro · 2 sellers · $1.00–$2.00 per million input tokens
  • Grok 4.6 · 2 sellers · $2.00–$2.20 per million input tokens
  • Qwen3 Next 80B A3B Thinking · 2 sellers · $0.150–$0.150 per million input tokens

Excluded from the ranked list

  • 0 unpriced — no finite input price (missing price is not free).
  • 206 single-seller — one shop is not a market.
  • 17 no actionable equal-or-better spread.
  • 31 lower or unknown precision — listed above, not as savings.
  • 0 where the cheaper input seller is not, or cannot be shown to be, cheaper on the balanced blend.
  • 49 no declared serving precision across sellers.
  • 0 on the publication denylist — never shown.

Questions this page answers

How is spread computed on this page?

Spread is dearest ÷ cheapest input price among distinct sellers at the reference serving precision or better. A vendor listing the same model at its own service tiers is not competition. A spread is listed only when the cheaper seller is also cheaper on the balanced blend the site ranks by (3 parts input to 1 part output); otherwise the input spread would point at the dearer seller.

Why are some models excluded?

Unpriced models are excluded. Cheaper endpoints at lower precision do not count as a saving if the reference precision is higher.

Does a high spread mean I should switch?

It means multiple sellers publish different rates at equal-or-better precision. Capability dominance is on /check/ and /frontier/; this page is shopping geometry only.

Input hashes & method version

Read citation text and BibTeX
45 models have actionable spreads. Condition: equal-or-better declared precision; cheaper lower-precision is not a saving; the cheaper input seller is also cheaper on the balanced blend (3 input : 1 output). Prices are USD per million input tokens. As of 2026-10-06. Lens: Published provider prices in the accepted OpenRouter-derived catalogue. https://undominated.ai/spreads/
Method: https://undominated.ai/methodology/
Publication: sha256:ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932; calculation SHA-256: 5ae4d0bee9b70298dd437b05df87c32acb407040b50bf88240677f295f25955d; permitted inputs: https://undominated.ai/data/publications/manifests/ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932.json

@misc{undominated-spreads-2026-10-06,
  author = {{Undominated.ai}},
  title  = {spreads evidence — Undominated.ai},
  year   = {2026},
  url    = {https://undominated.ai/spreads/},
  note   = {as of 2026-10-06; Published provider prices in the accepted OpenRouter-derived catalogue},
  annote = {Publication: sha256:ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932; calculation SHA-256: 5ae4d0bee9b70298dd437b05df87c32acb407040b50bf88240677f295f25955d; permitted inputs: https://undominated.ai/data/publications/manifests/ecef83f767cd4d2c21766f6a5b4dffa36c06351f23657542d473bc1a030c0932.json}
}
Evidence & Ask
Markdown for LLMs