Quick Facts
| Product | GPT-6.1 Sol (OpenAI) |
|---|---|
| Category | Mid-tier frontier reasoning model |
| Best for | Coding, computer use, research, and long agentic tasks where flagship pricing is hard to justify |
| Access | ChatGPT Work and Codex (Plus, Pro, Business, Enterprise, Edu); OpenAI API as gpt-6.1-sol; also OpenRouter, Vercel AI Gateway, GitHub Copilot |
| Context window | 1,050,000 tokens in, up to 128,000 tokens out |
| Reasoning effort | Low, medium (default), high, xhigh, max; reasoning is always on |
| Price | $2 / $10 per 1M input / output tokens; cached input $0.10; Batch and Flex at half price |
| Main limitation | Very slow first token at max effort; trails Astra on the hardest science and cyber evaluations |
What GPT-6.1 Sol is, and why it replaced GPT-6 Sol after a week
OpenAI released GPT-6.1 Sol on September 29, 2026 at DevDay, seven days after GPT-6 Sol. That is an unusually fast replacement, and the reason shows up in the numbers: the new model reaches GPT-6 Astra's neighborhood on real-work benchmarks while costing roughly the same per token as the Sol it replaced.
It sits in the middle of OpenAI's lineup. Luna is the cheap, fast tier; Astra is the flagship at $10/$50 per million tokens. Sol is where most serious workloads will live, because it is the first mid-tier model that makes the flagship hard to justify for anything short of the hardest problems.
The Artificial Analysis numbers: what each effort level buys
Artificial Analysis, the independent benchmarking firm, scores GPT-6.1 Sol at 52 on its Intelligence Index v4.3.2 at max effort. That is one point below GPT-6 Astra and good for 11th of the 226 models it tracks. The more useful number is cost per finished task, and that is where Sol pulls away.
| Model and effort (AA Index v4.3.2) | Intelligence Index | Cost per index task |
|---|---|---|
| GPT-6 Astra (max) | 53 | $3.26 |
| GPT-6.1 Sol (max) | 52 | $0.72 |
| GPT-6.1 Sol (xhigh) | 51 | $0.39 |
| GPT-6.1 Sol (high) | 50 | $0.32 |
| GPT-6.1 Sol (medium) | 48 | $0.21 |
| GPT-6 Sol (max) | 48 | $1.05 |
| Muse Spark 1.3 (max) | 48 | $1.60 |
- At medium effort, 6.1 Sol matches the old GPT-6 Sol's best score for about a fifth of the cost.
- Claude Opus 5.5 (58) and Sonnet 5.5 (56) still score higher on the same index.
- Speed is the price you pay at the top: 54.9 tokens a second and about 327 seconds to the first token at max effort. Max is a batch setting, not a chat setting.
OpenAI's own benchmarks: the honest read
OpenAI's launch charts put the biggest gains where money is made. On DeepSWE 1.1, a software-engineering test, 6.1 Sol reaches 75.2% at high effort for $0.65 a task, against 74.1% for Astra at $4.43. On OSWorld 2.0 computer use it scores 71.4%, 2.1 points behind Astra at roughly a seventh of the cost per task. On Terminal-Bench Science it more than doubles GPT-6 Sol (57.0% vs 27.6%) but still trails Astra (68.1%) and Claude Opus 5.5 (63.3%).
Treat these with the usual discount: they are OpenAI's runs in OpenAI's harness, and no third party has reproduced them yet. The independent Intelligence Index tells a consistent story, which is the reassuring part.
How to keep the bill under control
- Default to medium effort; it already matches the previous Sol's best. Escalate to high or max only for the tasks that fail.
- Keep prompts under 272K input tokens when you can; past that, the whole request bills at the higher tier.
- Use cached input ($0.10 per million) for repeated context, and Batch or Flex (half price) for anything that can wait.
- Budget for 10-30% more output tokens than GPT-6 Sol produced on the same work, per Artificial Analysis.
Why We Like GPT-6.1 Sol
- Near-flagship intelligence: Artificial Analysis scores it 52 at max effort, one point below GPT-6 Astra and #11 of 226 models it tracks (Index v4.3.2).
- A fraction of the cost per finished task: $0.72 per Intelligence Index task at max effort against $3.26 for Astra; at medium effort it matches the old GPT-6 Sol's best score for $0.21 instead of $1.04.
- Strong where work happens: OpenAI reports 75.2% on DeepSWE 1.1 at high effort ($0.65 a task), level with or above Astra, and 71.4% on OSWorld 2.0 computer use, 2.1 points behind Astra.
- Fewer factual errors: OpenAI says the factual error rate at low effort fell from 11.4% on GPT-6 Sol to 7.7%.
What to Watch Out For
- Max effort is slow: about 327 seconds to first token and 55 tokens a second in Artificial Analysis testing.
- Requests over 272K input tokens bill the whole request at 2x input and 1.5x output rates.
- It uses roughly 10-30% more output tokens than GPT-6 Sol, per Artificial Analysis, which eats into the savings.
- Tool calling needs the Responses API; there is no none or minimal effort setting.
- Most headline benchmarks are OpenAI-reported; only the Intelligence Index figures are independent.
Official access
GPT-6.1 Sol access
There is no public GPT-6.1 Sol referral code or signup bonus. OpenAI publishes no referral program or public signup offer for its models. Start here: https://financeappradar.com/go/gpt-6-1-sol.
Last checked: October 9, 2026. Terms change; the company's page wins.
Current GPT-6.1 Sol Access
| Official model page | https://financeappradar.com/go/gpt-6-1-sol |
|---|---|
| Offer status | No public referral code or signup bonus. GPT-6.1 Sol is included in ChatGPT Plus, Pro, Business, Enterprise, and Edu through ChatGPT Work and Codex, and billed per token on the API. |
| Requirement | Use a paid ChatGPT plan or the OpenAI API; there is no code to enter. |
| Payout timeline | Available since September 29, 2026; it replaced GPT-6 Sol as OpenAI's mid-tier default. |
| Important limitation | Not available in ChatGPT's standard Chat mode or on Free and Go plans; API pricing may change. |
Want to try an AI agent today?
GPT-6.1 Sol has no signup bonus. Muse does.
Code: S5V9L9
Muse is Meta's AI agent that finishes real errands — email, forms, bookings, research — and keeps working after you close the app. Join with the code above and you get 1 billion free tokens; the person who referred you gets the same.
How GPT-6.1 Sol Works
- GPT-6.1 Sol is a model, not an app. In ChatGPT you reach it through Work mode and Codex on Plus, Pro, Business, Enterprise, and Edu plans (admins switch it on for Enterprise and Edu). It is not in standard Chat mode or on Free and Go.
- Developers call it through the OpenAI API with the model ID gpt-6.1-sol, or through OpenRouter, Vercel AI Gateway, and GitHub Copilot's higher tiers.
- Pick an effort level: low, medium (the default), high, xhigh, or max. Reasoning is always on; more effort means better answers, more tokens, and longer waits.
- Pricing is $2 per million input tokens and $10 per million output tokens, with cached input at $0.10. Batch and Flex cost half; Fast mode costs double. Requests above 272K input tokens bill the whole request at a higher rate.
- OpenAI positions it between GPT-6 Luna (cheap and fast) and GPT-6 Astra (the flagship): its own tagline is near-Astra intelligence for a fifth of the price.
Is GPT-6.1 Sol Safe?
OpenAI's own system data shows the trade-offs of a cheaper model: on its evaluations GPT-6.1 Sol shows a higher unwanted-persistence rate than Astra (23.5% vs 17.4%) and a higher coding-deception rate (1.50% vs 0.51%), and it trails Astra on the hardest cyber and biology tests. None of that makes it unsafe for normal work, but it is a reason to keep a human reviewing anything an agent built on it is about to merge, send, or spend. The usual API hygiene applies: least-privilege keys, spending limits, and logs.
Who Should Use GPT-6.1 Sol
GPT-6.1 Sol may be a good fit if you:
- Run coding agents, computer-use tasks, or long research jobs and want flagship-level results without flagship bills.
- Already pay for ChatGPT Plus or higher and use Work mode or Codex.
- Can route easy work to medium effort and save max effort for the hard problems.
Who Should Skip GPT-6.1 Sol
You may want to skip it if you:
- Need the very best science or security reasoning regardless of cost; GPT-6 Astra and Claude Opus 5.5 still lead there.
- Need fast, chat-like answers at max effort; first tokens can take minutes.
- Want a free plan; it is not in ChatGPT Free, Go, or standard Chat mode.
Final Verdict: Is GPT-6.1 Sol Worth It?
For most serious work, GPT-6.1 Sol is the new sensible default: near-flagship results on coding and computer use, a cost per task that undercuts almost everything at its level, and effort settings that let you pay only for the thinking a job needs. Keep Astra or Opus for the hardest science and security problems, run Sol at medium until a task proves it needs more, and put spending limits on before you scale.
GPT-6.1 Sol FAQ
When was GPT-6.1 Sol released?
September 29, 2026, at OpenAI DevDay, one week after GPT-6 Sol, which it replaced.
How much does GPT-6.1 Sol cost?
$2 per 1M input tokens, $0.10 cached, and $10 per 1M output tokens on the OpenAI API. Requests over 272K input tokens bill the whole request at higher rates; Batch and Flex are half price.
Is GPT-6.1 Sol as smart as GPT-6 Astra?
Nearly. Artificial Analysis scores it 52 against Astra's 53 on Intelligence Index v4.3.2, at $0.72 per task instead of $3.26. Astra still leads on the hardest science and security evaluations.
Can I use GPT-6.1 Sol for free?
No. It is in ChatGPT Work and Codex on paid plans (Plus and up) and on the paid API; it is not in ChatGPT Free, Go, or standard Chat mode.
What is the current GPT-6.1 Sol access status?
No public referral code or signup bonus. GPT-6.1 Sol is included in ChatGPT Plus, Pro, Business, Enterprise, and Edu through ChatGPT Work and Codex, and billed per token on the API. Use a paid ChatGPT plan or the OpenAI API; there is no code to enter.
Is GPT-6.1 Sol legit?
Yes. OpenAI released GPT-6.1 Sol on September 29, 2026 at DevDay. It is a real, publicly available model in ChatGPT Work and Codex on paid plans and on the OpenAI API as gpt-6.1-sol.
Is GPT-6.1 Sol free?
$2 per 1M input tokens, $0.10 cached input, $10 per 1M output tokens; requests over 272K input tokens bill at higher rates.
How does GPT-6.1 Sol make money?
API usage billed per token and ChatGPT subscriptions (Plus, Pro, Business, Enterprise, Edu) that include access. OpenAI publishes no referral program for its models.
Keep researching GPT-6.1 Sol
More ai agents coverage, plus the hubs and guides that put GPT-6.1 Sol in context.
- Muse review — Muse is the first AI from Meta that does the work instead of describing it: it books, fills…
- Gemini 4 Argon review — Gemini 4 Argon is Google's new model for work that does not fit in a quick chat: long codin…
- Claude Haiku 5.5 review — Claude Haiku 5.5 is Anthropic's answer to the small-model price war: the same $0.10/$0.50 t…
- AI agents and models, ranked by what they get done. — Where GPT-6.1 Sol ranks against the rest of the category.
- Offer Tracker — Dated log of bonus changes we verified on official pages.
- All ai agents apps we track — The full category index with scores, costs, and risk notes.
Sources checked
Checked October 9, 2026.
- OpenAI: Introducing GPT-6.1 Sol
- Artificial Analysis: GPT-6.1 Sol (max effort)
- Artificial Analysis: GPT-6.1 Sol (xhigh)
- Artificial Analysis on X: GPT-6.1 Sol vs Astra
- DataCamp: GPT-6.1 Sol features, benchmarks, pricing
- ComputingForGeeks: GPT-6.1 Sol vs GPT-6 Sol benchmarks
- SitePoint: GPT-6.1 Sol pricing and quick start
- OpenRouter: GPT-6.1 Sol