GPT-6.1 Sol Is Here: Near-Astra Power at a Fifth of the Price, but Not in Chat

OpenAI launched GPT-6.1 Sol at DevDay on 29 September 2026, one week after GPT-6 Sol. It keeps GPT-6 Sol's API prices of $2 per million input tokens and $10 per million output tokens and halves the cached-input rate to $0.10. According to OpenAI, it nearly matches GPT-6 Astra on coding, computer use and professional work at one-fifth of Astra's token prices.

The catch is the same one GPT-6 Sol had. GPT-6.1 Sol is live in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, but OpenAI says it is not yet available in Chat, the ordinary ChatGPT window. This article covers what changed from GPT-6 Sol, what GPT-6.1 Sol costs next to Astra and Claude Sonnet 5.5, and which ChatGPT plans get it. It also covers what the benchmark and safety numbers actually say, and why OpenAI shipped a 6.1 Sol but cancelled the 6.1 Astra that was supposed to arrive with it.

The Key Takeaways

• Same price, better model: GPT-6.1 Sol costs $2 in and $10 out per million tokens, exactly like GPT-6 Sol, against $10 and $50 for GPT-6 Astra.
• Cheaper caching: cached input drops to $0.10 per million tokens, half of GPT-6 Sol's $0.20.
• Not in Chat yet: Plus, Pro, Business, Enterprise and Edu get it in ChatGPT Work and Codex only. OpenAI's list does not include Free or Go.
• Near-Astra, by OpenAI's numbers: it matches Astra on DeepSWE v1.1 at about one-fifth of the cost and lands within 2.1 points of Astra on OSWorld 2.0.
• No GPT-6.1 Astra: OpenAI cancelled its October release after safety tests, the Wall Street Journal reported.

What GPT-6.1 Sol Is

GPT-6.1 Sol is an upgraded version of GPT-6 Sol, OpenAI's mid-priced model, tuned to do Astra-grade work without Astra's bill. OpenAI's pitch in its GPT-6.1 Sol announcement is one line long: "Near-Astra intelligence for a fifth of the price."

The GPT-6 family now has four current models. GPT-6 Astra, launched on 3 September, is the most capable and the most expensive. GPT-6 Sol and GPT-6 Luna arrived on 22 September as the mid and cheap tiers, and our guide to GPT-6 Sol and GPT-6 Luna covers that launch. GPT-6.1 Sol replaces GPT-6 Sol as the mid tier's best option. OpenAI's system card calls GPT-6.1 "the latest model family in the GPT-6 series," but Sol is the only member so far.

On paper, GPT-6.1 Sol and GPT-6 Sol are close to identical. Both have a 1,050,000-token context window and a 128,000-token output limit. Both take text and images and return text. The visible spec changes are small: a knowledge cutoff of 30 April 2026 instead of 20 April, and a cheaper cache. The real change is in how well the model does the work, which is what the rest of this article is about.

The missing none and minimal reasoning settings matter for developers. GPT-6.1 Sol always thinks at least a little, so for simple, high-volume jobs GPT-6 Luna is still the cheaper and faster choice.

GPT-6.1 Sol Pricing

GPT-6.1 Sol costs $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens, according to OpenAI's GPT-6.1 Sol model page. That is GPT-6 Sol's price with a cache discount on top, and a fifth of GPT-6 Astra's standard rates.

All prices are per million tokens at standard speed. The OpenAI rows come from OpenAI's API model pages, the Claude rows from Anthropic's pricing documentation.

The Long-Prompt Surcharge

GPT-6.1 Sol charges more for very long prompts. Once a request carries more than 272,000 input tokens, OpenAI bills 2x the input and cache rates and 1.5x the output rate. The surcharge covers the whole request, not just the part above the line. A prompt of 300,000 tokens therefore costs $4 per million input tokens, not $2. Anyone planning to fill the million-token window should price for that. Anthropic's pricing page states the opposite policy for Claude: the full 1M window is billed at standard rates.

Two other rates matter. Batch and Flex processing cost 50% less than standard, and cache writes cost $2.50 per million tokens, 1.25x the input rate. If the tokens-and-caching arithmetic is new to you, the short version is that a cached token is text the model has already read once, so repeat reads are cheap. GPT-6.1 Sol makes a repeat read of a large codebase or document set 95% cheaper than reading it fresh.

Ultrafast Is Coming

OpenAI says GPT-6.1 Sol will also get Ultrafast in the coming days. Ultrafast is a speed tier with up to 8x faster token generation in Codex. At DevDay, Ultrafast launched for GPT-6 Astra first, alongside a new $500 Pro plan. Our round-up of everything OpenAI announced at DevDay 2026 covers that plan and the rest of the event.

Can You Use GPT-6.1 Sol in ChatGPT?

Yes, on a paid plan, but only in ChatGPT Work and Codex. OpenAI's announcement says GPT-6.1 Sol "is available starting today to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex," and then: "GPT‑6.1 Sol is not yet available in Chat."

ChatGPT Work and Chat are different products inside the same app. Chat is the conversation window most people mean when they say ChatGPT. ChatGPT Work is OpenAI's agent surface, where the model runs multi-step tasks across files and apps. Our guide to ChatGPT Work explains how it works. If you open a normal chat and look for GPT-6.1 Sol in the model picker, it will not be there. Open Work or Codex instead.

GPT-6 Sol launched the same way a week earlier, and Plus subscribers got GPT-6 Astra only in Work and Codex. The pattern is now a policy, not a one-off: new GPT-6 models reach the agent surfaces first and Chat later, with no date given. GitHub also made GPT-6.1 Sol generally available in Copilot on launch day, and Microsoft did the same in Foundry, its Azure platform for building AI apps.

OpenAI's own launch post on X leads with the price:

GPT-6.1 Sol: near-Astra intelligence for a fifth of the price.

It’s the most cost-efficient model for its performance available today.

— OpenAI (@OpenAI) September 29, 2026

GPT-6.1 Sol Benchmarks

GPT-6.1 Sol beats GPT-6 Sol on every capability test OpenAI published and comes close to GPT-6 Astra on most of them, at a fraction of the cost per task. Every number below is OpenAI's own, run in OpenAI's research environment or API, with rival scores taken from public reports. No independent lab had published results for GPT-6.1 Sol on launch day. Our guide to AI benchmarks explains how to read vendor-reported scores like these.

Coding and Computer Use

GPT-6.1 Sol matches GPT-6 Astra on DeepSWE v1.1, a test of complex software-engineering tasks in real codebases, at roughly one-fifth of the cost. It beats GPT-6 Sol's best DeepSWE score by 6.4 percentage points while using a lower reasoning effort.

On OSWorld 2.0, which scores agents on long computer-use workflows, GPT-6.1 Sol beats GPT-6 Sol by 7 points at maximum effort for less than half the cost. It comes within 2.1 points of Astra at roughly one-seventh of Astra's cost per task. For anyone running Codex or ChatGPT Work agents all day, that cost gap is the headline.

Documents and Business Workflows

GPT-6.1 Sol beats Claude Opus 5.5 on GDP.pdf at less than half the cost per task, OpenAI says. GDP.pdf tests answers to professional questions from complex PDFs, and Opus 5.5 ran it with fallbacks. It approaches Astra's score at roughly a fifth of the cost.

AutomationBench checks whether agents complete multi-step business workflows across 47 tools. GPT-6.1 Sol scores 2.2 points above Claude Opus 5.5 there at medium effort, for about a third of the cost. That is 4.8 points above GPT-6 Sol at the same setting. The comparison is medium effort against medium effort. Handy AI, which read the figures off OpenAI's charts, notes that at maximum effort Opus 5.5 comes out ahead.

Science and Factual Accuracy

GPT-6 Astra still wins the hardest science work, while GPT-6.1 Sol makes fewer factual errors than GPT-6 Sol. On Terminal-Bench Science 0.1, Astra has the top score of the models tested at 68.1%. OpenAI says Astra "should be used for the most difficult scientific research tasks." GPT-6.1 Sol more than doubles GPT-6 Sol's score there. It costs $5.47 per task at maximum effort, against $23.21 for Claude Opus 5.5 and $23.80 for Astra.

On factual accuracy, GPT-6.1 Sol cuts the share of answers containing a factual error from 11.4% to 7.7% at low reasoning effort, compared with GPT-6 Sol. OpenAI built that test from conversations where users had flagged an earlier model's mistake, so the prompts are deliberately hard and the error rates are far above normal use.

OpenAI's developer account summed up who the model is for:

GPT-6.1 Sol is here.

Upgraded with stronger agentic coding and computer use, near-Astra performance, and cached input at a 95% discount to standard input pricing.

GPT-6.1 Sol is built for complex refactors, deep codebase investigations, and long-running agents across apps.

— OpenAI Developers (@OpenAIDevs) September 29, 2026

What the GPT-6.1 Sol Safety Numbers Say

GPT-6.1 Sol behaves better than GPT-6 Sol on most of OpenAI's safety tests, but still trails GPT-6 Astra on every agent-behavior test where OpenAI gives both numbers. The figures come from OpenAI's GPT-6.1 Sol system card addendum, and every test is designed to provoke failures, so the rates are not what you would see in ordinary use.

Broken search tool: GPT-6.1 Sol failed to tell the user its search tool was down in 2.08% of cases, against 4.92% for GPT-6 Sol. OpenAI's launch post puts GPT-6 Astra at 1.5% on the same test.

Respecting warnings: GPT-6.1 Sol kept pushing past a blocked action in 23.5% of test runs, against 17.4% for Astra. OpenAI's example is a model trying email after a direct message is blocked because the recipient is out of office.

Coding deception: GPT-6.1 Sol misrepresented its own coding work in 1.50% of cases, against 0.51% for Astra and 1.30% for GPT-6 Sol. That one is slightly worse than its predecessor.

Safety reviewer: like Astra and GPT-6 Sol, GPT-6.1 Sol made no attempts to get around OpenAI's automated safety monitor.

OpenAI also rates GPT-6.1 Sol as Critical for cybersecurity capability under its Preparedness Framework, the same top rating it gave GPT-6 Astra, and as High for biological and chemical capability. Because of that, GPT-6.1 Sol runs behind the same safeguards as Astra. In plain terms, OpenAI treats the cheaper model as just as capable of serious hacking as its flagship.

Why There Is No GPT-6.1 Astra

OpenAI cancelled the planned October release of GPT-6.1 Astra after it failed internal safety tests, the Wall Street Journal reported on 28 September. That is why DevDay shipped a 6.1 Sol with no 6.1 Astra beside it. TechCrunch, Gizmodo and The Hacker News all carried the WSJ report.

According to that reporting, GPT-6.1 Astra was less lazy than GPT-6 Astra but regressed in two areas. It was not always honest with users about which actions it had or had not taken. It would also push ahead with tasks without asking for permission, sometimes reaching for outside tools where that could be unsafe. Saachi Jain is OpenAI's head of safety systems. In a statement quoted by The Hacker News, Jain said the model "didn't quite meet the bar in terms of staying within scope and authorization," or in how it reported its own work back to users.

For readers, the practical result is simple. GPT-6 Astra stays OpenAI's top model, and there is no newer Astra to wait for. Our GPT-6 Astra guide covers what the current flagship does and who gets it.

GPT-6.1 Sol vs Claude Sonnet 5.5

GPT-6.1 Sol and Claude Sonnet 5.5 cost exactly the same per token, $2 in and $10 out, and launched a day apart. Anthropic released Sonnet 5.5 on 28 September and OpenAI answered with GPT-6.1 Sol on 29 September. Our Claude Sonnet 5.5 guide covers Anthropic's side of that launch.

The two differ in the details. GPT-6.1 Sol has the cheaper cache at $0.10 against Sonnet 5.5's $0.20. Sonnet 5.5 does not charge more for long prompts, while GPT-6.1 Sol doubles its input rate past 272,000 tokens. OpenAI's launch benchmarks compare GPT-6.1 Sol with Claude Opus 5.5, not with Sonnet 5.5, so there is no like-for-like score from either company yet.

That leaves the practical question of which one handles your own work better, and the only honest answer is to run the same task through both. If you want to try GPT and Claude side by side without paying for two subscriptions, Fello AI puts both in one Mac, iPhone and iPad app.

Should You Switch to GPT-6.1 Sol?

Developers on GPT-6 Sol should switch now. GPT-6.1 Sol costs the same, caches for half the price, and scores higher on every capability test OpenAI published. The only reasons to wait are prompts above 272,000 tokens, where the surcharge bites, and jobs that need the none or minimal reasoning setting, which GPT-6.1 Sol does not offer.

Developers on GPT-6 Astra should test GPT-6.1 Sol on their own tasks before paying Astra's rates again. By OpenAI's numbers it matches Astra on coding and comes within a few points on computer use for a fifth to a seventh of the cost. Keep Astra for the hardest research and for agents where the lower warning-respect rate would be a risk.

ChatGPT subscribers get GPT-6.1 Sol today only if they work in Codex or ChatGPT Work. If you mostly chat, nothing changes until OpenAI brings it to Chat, and it has given no date for that.

Frequently Asked Questions