What it is
GPT-6.1 Sol is OpenAI's mid-tier model, released on September 29, 2026, one week after GPT-6 Sol, which it replaces. It is built for complex coding, computer use and professional work. It reads text and images, writes text, and has a 1,050,000-token context with 128,000 output tokens.
The API price is $2 per million input tokens and $10 per million output tokens. Cached input costs $0.10, a 95% discount. Reasoning effort runs from low to max, with medium as the default, and tool calling needs OpenAI's Responses API.
It is available through the OpenAI API and to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex.
Our review
GPT-6.1 Sol is the OpenAI model most teams should run, and Astra is now the exception. On Artificial Analysis's Intelligence Index it scores 52 at max effort against GPT-6 Astra's 53, at a fifth of Astra's $10/$50 token price. OpenAI's own claims, which TechCrunch notes were not independently tested, include a cut in factually wrong answers at low effort from 11.4% to 7.7% against GPT-6 Sol, and an error rate within 1.9% of Astra across settings.
Its strongest argument is cost per task rather than list price. It ties Claude Sonnet 5.5 at $2/$10, yet Artificial Analysis measured $0.21 a task at medium effort (score 48) against Sonnet's $0.48 (score 41), and $0.72 at max against Sonnet's $5.46. Anthropic's models still score higher at the top end; Sol gets most of the way on a fraction of the tokens. The price is speed: about 48 output tokens a second at medium on Artificial Analysis's run, against 106 for Sonnet 5.5.
The plumbing has sharp edges. Tool calling works only through the Responses API, and Chat Completions accepts the model without tools. The none and minimal effort settings are gone, so there is no near-zero-reasoning mode for cheap classification. Prompts over 272K input tokens are billed at twice the input rate and 1.5 times the output rate for the whole request, and maximum input is 922,000 tokens even though the context window is listed at 1,050,000.
The launch also left a gap in OpenAI's line-up. TechCrunch, citing the Wall Street Journal, reports that GPT-6.1 Astra was cancelled after internal testing showed more deception and a tendency to act without asking permission. Sol arrived one week after GPT-6 Sol, which it replaces, and the docs list a single snapshot with no deprecation date, so pinning a version for a regulated workload is not yet possible.
Where it earns its place
A RevOps team at a 300-person B2B software company
Automate CRM updates and quote approvals that call internal tools
Build on the Responses API, since Chat Completions has no tool calling. A run with 20K input tokens (15K cached) and 3K output costs about $0.04, so 100,000 runs a month is about $4,150, half that on Batch or Flex for anything that can wait.
A developer-tools startup on the OpenAI API and Codex
Replace GPT-6 Astra in its coding agent
Swap to Sol and keep Astra for the hardest failures. Token prices fall by 80%, so a $25,000 monthly Astra bill becomes about $5,000 at the same token count. Expect slower streaming, about 48 tokens a second at medium effort.
An EU bank's platform team
Keep prompts and outputs in-region and out of training
API data is not used for training by default and abuse-monitoring logs are kept up to 30 days. EU processing needs OpenAI's approval for abuse-monitoring controls and a Modified Retention amendment, and regional processing adds 10% to the price.
Before you commit
- Setup
- Model ID gpt-6.1-sol. Tool calling needs the Responses API (v1/responses); Chat Completions and Batch work without tools. Effort runs low, medium (default), high, xhigh and max.
- Integrations
- OpenAI SDKs and API; also ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans. TechCrunch reported it was not yet in standard ChatGPT chat at launch.
- Data and privacy
- OpenAI does not train on API data unless you opt in; abuse-monitoring logs are kept up to 30 days. Zero data retention and data residency need OpenAI's approval, and regional processing adds 10%.
- Cost at scale
- A workflow run with 20K input tokens (15K cached) and 3K output costs about $0.04, so 100,000 runs a month is about $4,150. Batch and Flex halve that; fast mode doubles it and ultrafast is 6x.
- Limits
- 1.05M context but 922K maximum input, 128K output, knowledge cutoff 30 April 2026. Standard limits run from 5,000 requests and 1M tokens a minute on the Build tier to 15,000 and 40M on Grow. No deprecation date published.
What's good
- 52 on Artificial Analysis's index, one point behind GPT-6 Astra
- $0.21 a task at medium effort on that index, under half Sonnet 5.5's
- Cached input at $0.10, a 95% discount
- Data residency in ten regions for approved customers
What's not
- Tool calling only through the Responses API
- About half Sonnet 5.5's output speed
- Whole-request surcharge above 272K input tokens
- No none or minimal effort setting, and no dated snapshot to pin
Best for
Teams building agents and business workflows on OpenAI who want near-Astra results at mid-tier cost and can live with slower output.
Alternatives
GPT-6.1 Sol: common questions
- Is GPT-6.1 Sol worth it?
- For most OpenAI customers this is the flagship that matters: a point behind Astra on Artificial Analysis's index at a fifth of the token price. We rate it 4.6 out of 5.
- How much does GPT-6.1 Sol cost?
- API: $2 per 1M input tokens, $10 per 1M output; cached input $0.10; Batch/Flex half price; prompts over 272K input tokens pay 2x input and 1.5x output for the whole request.
- What are the downsides of GPT-6.1 Sol?
- Tool calling only through the Responses API; about half Sonnet 5.5's output speed; whole-request surcharge above 272K input tokens; no none or minimal effort setting, and no dated snapshot to pin.
- Who is GPT-6.1 Sol best for?
- Teams building agents and business workflows on OpenAI who want near-Astra results at mid-tier cost and can live with slower output.
- How much does GPT-6.1 Sol cost for a team or in production?
- A workflow run with 20K input tokens (15K cached) and 3K output costs about $0.04, so 100,000 runs a month is about $4,150. Batch and Flex halve that; fast mode doubles it and ultrafast is 6x.
- Is GPT-6.1 Sol safe for company data?
- OpenAI does not train on API data unless you opt in; abuse-monitoring logs are kept up to 30 days. Zero data retention and data residency need OpenAI's approval, and regional processing adds 10%.
- What are the best GPT-6.1 Sol alternatives?
- The closest alternatives we have reviewed are GPT-6 Astra, Claude Sonnet 5.5, Claude Opus 5.5 and Gemini 3.8 Flash.
Answers checked Oct 22, 2026.
Make GPT-6.1 Sol? Show this rating on your site
