Model highlight

GPT-6.1 Sol for Chemical Companies: Near the Top for an Eighth of the Bill

OpenAI's GPT-6.1 Sol scores 10 points under Claude Fable 5.2, the leader of the Artificial Analysis Intelligence Index, and a few points under Gemini 4 Carbon and Claude Opus 5.5, and runs one of its tasks for $1.27 where those three cost $26.54, $5.48, and $10.04. For a chemical distributor or producer, that makes it a candidate for research you run often, with limits to test first.

54.6on the Artificial Analysis Intelligence Index, at $1.27 a task
OpenAI · Released 29 Sep 2026

Latest update

vals.ai added Google's Gemini 4 Neon at 71.21% on regulatory research, now the leader, so GPT-6.1 Sol's 53.03% sits just outside the leader's margin, by 1.47 points, instead of just inside it. Sol is 33rd of 101.

1 of 13

GPT-6.1 Sol at a glance

Made by
OpenAI7
Released
29 Sep 20261
Price
$2 per million tokens sent, $10 per million written7
Reads
Text and images, up to 1,050,000 tokens7
Settings
Five, from low to max; medium is the default7
Knows up to
30 Apr 20267
Index
54.6, well above the 38 median for models at its price2

What GPT-6.1 Sol is.

OpenAI released GPT-6.1 Sol on 29 September 2026. OpenAI describes it as “near-Astra performance for complex work at a lower cost”, where Astra is its larger GPT-6 model. It reads text and images, holds up to 1,050,000 tokens, and costs $2 for every million tokens you send and $10 for every million it writes.

It is a general model: it drafts, summarizes, and reasons through research. You choose how hard it thinks, from low to max. The scores on this page are at max, the setting each test ran.

Near the top, for an eighth of the bill.

Each dot is a model on the Artificial Analysis Intelligence Index: how high it scores, and what one of the index's tasks costs to run on it. Sol sits about 10 points under Claude Fable 5.2 at the top and a few points under Gemini 4 Carbon and Claude Opus 5.5, at a fraction of their cost. The line under the chart sets one model against Sol; tap any dot to change it.

Index score
10203040506070$0.10$0.30$1$3$10$30
Cost of one index task, US dollars
GPT-6.1 SolOpen weights, run it yourselfClosed model

Claude Opus 5.5: 59.1 on the index, 4.5 points above Sol, at $10.04 a task, 7.9 times Sol's cost. Tap any dot to compare.

Artificial Analysis: models and the Intelligence IndexNo date stated

Where Sol lands on the guide's tests.

Two of the guide's jobs have an independent test that ran GPT-6.1 Sol, both at its max setting.

US regulatory research questions answered fully right, out of 66

  1. Gemini 4 Neon, the leader71.21%Tie
  2. GLM-5.3 Flash, $0.199 a question57.58%Tie
  3. Muse Ember, $0.056 a question, the last inside the margin54.55%Tie
  4. GPT-6.1 Sol, $2.76 a question53.03%
  5. DeepSeek V4.1 Flash, $0.25 a question53.03%

Sol now sits just outside the leader's margin of error, by 1.47 points, since Gemini 4 Neon took first place at 71.21%. Muse Ember, at 54.55% for $0.056 a question, is the last result the test cannot separate from the leader. Three other models score exactly 53.03%, and one of them costs about a tenth as much.

vals.ai Legal Research, regulatory questions19 Nov 2026

Where Sol falls short.

A smaller bill buys a smaller result in some places. These are the ones the tests show.

  • Finished Work Trails the Leaders

    On rated professional work it is 19th of 29, behind open models such as DeepSeek V4.1 Flash 1017 and MiMo V2.6 Pro. For a report a customer reads, the guide's pick is a larger model.

    Artificial Analysis: models and the Intelligence IndexNo date stated

  • Its Regulatory Result Sits Just Outside the Top Group

    53.03% now sits just outside the leader's margin, by 1.47 points, on 66 questions of US law. Three other models score exactly the same, one of them for $0.25 a question.

    vals.ai Legal Research, regulatory questions19 Nov 2026

  • One Document Test, at Its Low Setting

    ParseBench, run by a company that sells a rival tool, scored GPT-6.1 Sol only at its low setting: 69.63 overall, 46th of 157, and 9th on tables. OmniDocBench has not scored it. Do not assume either result carries over to your price lists or certificates.

    ParseBench leaderboard14 Nov 2026

  • Gemini 4 Argon's and Carbon's Prices Double on 1 December

    Google now offers Argon and Carbon to every paying customer, at $2 per million tokens sent and $10 per million written through 30 November 2026. From 1 December it charges $4 and $20, so Carbon's $5.48 a task on the chart above will roughly double. Artificial Analysis has not scored Argon on its new index, so it is not on the chart above.

    Google: Gemini Developer API pricingNo date stated

  • Index Tasks Are Not Your Tasks

    The $1.27 is the average cost of the index's own tasks. Your cost depends on how long your documents are and how much Sol writes back.

    Artificial Analysis: models and the Intelligence IndexNo date stated

Questions about GPT-6.1 Sol.

Short answers, from the same sources as the rest of this page.

What is GPT-6.1 Sol?

A general model from OpenAI, released on 29 September 2026. OpenAI describes it as near-Astra performance at a lower cost. It costs $2 for every million tokens sent and $10 for every million written.

How good is GPT-6.1 Sol?

It scores 54.6 on the Artificial Analysis Intelligence Index, 10.0 points behind Claude Fable 5.2, which leads the index at 64.6.

Is GPT-6.1 Sol cheaper than Claude Sonnet 5.5?

Per token they cost the same: $2 in and $10 out. Per index task, Sol cost $1.27 and Sonnet 5.5 cost $10.02, because Sonnet used far more tokens. On regulatory research Sol cost $2.76 a question against $11.67.

Can a chemical company use GPT-6.1 Sol for regulatory research?

On a test of 66 US regulatory questions it got 53.03% fully right, 33rd of 101 and just outside the margin of the leader, Gemini 4 Neon. The test is US law, not REACH or GHS, so make it quote the dated official text, and check that text yourself.

Sources.

  1. Artificial Analysis: models and the Intelligence Index Artificial Analysis, an independent testing company. Each model's record gives its index score, its cost per index task, and its professional work rating.States no date · Vailent checked 16 Nov 2026
  2. Artificial Analysis: GPT-6.1 Sol Artificial Analysis, an independent testing company. It compares the model with others in a similar price tier.States no date · Vailent checked 3 Nov 2026
  3. Artificial Analysis: Gemini 4 Argon Artificial Analysis, an independent testing company, on its own tests of the model.Published 30 Sep 2026 · Vailent checked 5 Oct 2026
  4. Google: Gemini Developer API pricing The maker's own price list.States no date · Vailent checked 18 Oct 2026
  5. vals.ai Legal Research, regulatory questions vals.ai, an independent testing company. Lawyers wrote and checked the questions; an answer counts only if it meets every point on the lawyers' list.Published 19 Nov 2026 · Vailent checked 21 Nov 2026
  6. ParseBench leaderboard LlamaIndex, which sells LlamaParse, a product it ranks.Published 14 Nov 2026 · Vailent checked 15 Nov 2026
  7. OpenAI: GPT-6.1 Sol The maker's own page.States no date · Vailent checked 30 Sep 2026