GPT-5.6 Sol

OpenAI released 2026-07-09 proprietary

$4.00 per million tokens, balanced

Grok 4.6 scores higher and costs less — but you would give something up.

0 on the capability index and 25% cheaper ($3.00/M against $4.00/M). What you lose:

  • 1.1M → 500K context

Capability scores do not measure context length, output ceiling or which inputs a model accepts, so a higher score does not mean a drop-in replacement.

The same model via batch is 50% cheaper ($2.00/M) — same weights, different latency.

Our take

editorial — not a measurement

OpenAI's frontier tier and the clearest example of why a single headline price is misleading. Advertised at $4/$20, it silently becomes $8/$30 past the long-context threshold — and OpenAI, unlike xAI and Google, does not publish where that threshold is. It also carries promotional pricing described as holding only "at least through 2026-11-21", so the rate has a known expiry and no published successor. On measured intelligence it lands at AA 61, below Opus 5's 63, with a ~105s TTFT that is the worst in the leaderboard sample. Strong model, genuinely hard to budget for.

  • Unverified in this take: This note cites a time-to-first-token figure that the data layer withheld as implausible (it appears to be total reasoning time mislabelled at source). Treat it as unverified.

Strengths

  • AA intelligence index 61, competitive at the frontier
  • 75 output tokens/sec — faster generation than Opus 5
  • 1.05M context window
  • 50% batch discount and cheap cached input at $0.40

Weaknesses

  • Long-context tier doubles input to $8 and raises output to $30
  • Threshold for the long-context tier is not published
  • Promotional pricing with a stated ~2026-11-21 floor and no announced successor rate
  • ~105s time-to-first-token, worst in the sampled set
  • OpenRouter lists it at $2/$10, conflicting with OpenAI's own page

Reach for it when

  • Hard reasoning where OpenAI tooling is already in place
  • Short-context frontier tasks that stay under the tier threshold

Avoid it if

  • Your prompts are long and you need predictable cost
  • You need low latency
  • You are budgeting past November 2026

Sources: developers.openai.comartificialanalysis.aiarena.ai

Every price dimension

Input$2.00/M
Output$10.00/M
Cached input$0.200/M
Cache write$2.50/M
Web search$0.01/call
Batch discount50%

Past 272,000 tokens the price changes. Input goes to $4.00/M (2×) and output to $15.00/M. The headline rate does not apply to a long-context workload.

Independent scores

Intelligence60.9
Coding77.4
Agentic57.8
LMArena Elo1454.2xhigh

Sources are listed separately rather than averaged. Across the 62 models both have scored they correlate at r = 0.806 — close agreement overall, but four models rank very differently between them, and a blended score would hide exactly those.

Capability

Context window1.1M
Max output128K
Input modesfile, image, text
Tool useyes
Reasoningoptional
Knowledge cutoff2026-02-16
Open weightsno

Provenance

Price sourceopenrouter.ai
Fetched2026-08-24
Quality dataverified
Cross-checkedvendor page

A published time-to-first-token figure for this model was implausible (105s) and has been withheld rather than displayed.

GPT-5.6 Sol is the flagship model in OpenAI's GPT-5.6 series. It is suited for complex reasoning, coding, and agentic workflows, and is particularly strong at command-line and multi-step coding tasks...

Markdown for LLMs