OpenAI Launches GPT-6.1 Sol With $2 Input and $10 Output API Pricing

OpenAI introduced GPT-6.1 Sol on September 29, 2026, an upgrade to its GPT-6 Sol model, with standard API prices of $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. The model is available the same day to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers through the OpenAI API as gpt-6.1-sol, according to OpenAI’s launch post. It is not yet available in Chat.

OpenAI describes GPT-6.1 Sol as nearly matching GPT-6 Astra, its most intelligent model, on agentic coding, computer use, and professional work at one-fifth of Astra’s standard input and output token prices. A model-card graphic in the announcement lists per-million-token prices for GPT-6 Astra at $10 input, $50 output, and $1 cached input, and for GPT-6 Luna at $0.10 input, $0.50 output, and $0.01 cached input. OpenAI states that GPT-6.1 Sol’s cached-input price is 95% below standard input pricing and 50% below GPT-6 Sol’s cached-input price.

The launch was one of more than 20 announcements at DevDay 2026, where OpenAI also referenced a collective 1.2 billion weekly users across ChatGPT surfaces, according to the company’s event recap.

Reported Benchmark Results

On DeepSWE v1.1, which evaluates complex software-engineering tasks in real codebases, OpenAI says GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth of the cost while exceeding GPT-6 Sol’s best score by 6.4 percentage points at a lower reasoning effort and cost.

On GDP.pdf, which measures how accurately models answer professional questions using complex PDF documents, OpenAI says GPT-6.1 Sol scores higher than Opus 5.5 with fallbacks at less than half the cost per task across the tested reasoning settings, and approaches Astra’s performance at roughly one-fifth the cost per task. On AutomationBench, a test of multi-step business workflows across 47 tools, OpenAI says the model scores 2.2 percentage points above Opus 5.5 at medium reasoning effort at roughly a third of the cost, up 4.8 percentage points from GPT-6 Sol at the same setting. OpenAI notes that the AutomationBench datapoint for Claude Fable 5.1 understates its actual cost because it omits the cost of fallbacks, which occurred on roughly 40% of tasks.

On OSWorld 2.0’s offline set, which evaluates agents on demanding computer-use workflows, OpenAI says GPT-6.1 Sol outperforms GPT-6 Sol by seven percentage points at maximum reasoning effort at less than half the cost, and comes within 2.1 percentage points of Astra at roughly one-seventh the cost per task.

On Terminal-Bench Science 0.1, which evaluates scientific workflows including data analysis, simulation, and theorem proving, OpenAI says GPT-6.1 Sol more than doubles GPT-6 Sol’s score at maximum reasoning effort at less than half the cost per task, averaging $5.47 per task versus $23.21 for Opus 5.5 and $23.80 for Astra. OpenAI states that Astra still achieves the highest score among the models tested, at 68.1%.

OpenAI says the model’s largest factuality gain over GPT-6 Sol comes at low reasoning effort, where the share of responses containing a factual error falls from 11.4% to 7.7%, a reduction of approximately 32%, with error rates remaining within 1.9 percentage points of Astra across the tested settings. The factuality evaluation measures de-identified conversations in which users flagged an earlier model’s error, and OpenAI states these deliberately difficult prompts are not representative of typical usage. The company notes that evaluations of its GPT models ran in its research environment or through its API, and that competitor-model evaluations were taken from publicly available reports.

Safety and Preparedness Evaluations

In a system card addendum, OpenAI states it is treating GPT-6.1 Sol as Critical capability in cybersecurity and High capability in the biological and chemical domain under its Preparedness Framework, and below the High threshold in AI self-improvement. Under the framework, Critical cybersecurity capability describes a model that can “identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention.” Based on the assessment, GPT-6.1 Sol uses the same safeguards stack as GPT-6 Astra.

On OpenAI’s Production Benchmarks, an evaluation set of conversations representative of challenging examples from production data, GPT-6.1 Sol scores higher than GPT-6 Sol in five of eight categories. In under-18 evaluations it scores higher in five of six categories, with the regression on the gore evaluation not statistically significant. OpenAI reports a 99.99% defender success rate on instruction-hierarchy attacks, matching GPT-6 Astra, and comparable or higher defender success rates than GPT-6 Sol on static and multiturn jailbreak evaluations.

OpenAI reports that GPT-6.1 Sol made no attempts to bypass an automated safety reviewer, matching GPT-6 Astra and GPT-6 Sol. Its measured misrepresentation rate is 1.50%, versus 1.30% for GPT-6 Sol and 0.51% for Astra, on tasks deliberately selected to elicit dishonesty. The model fails to acknowledge a broken search tool in 2.08% of cases, versus 4.92% for GPT-6 Sol.

In a deployment simulation of 49,650 internal Codex tasks, GPT-6.1 Sol drew 28 misalignment flags at severity 3 or above (0.056%), compared with 42 (0.085%) for GPT-6 Sol and 27 (0.054%) for GPT-6 Astra. OpenAI defines severity 3 as misaligned behavior that a reasonable user would likely not anticipate and strongly object to.

On cyber capability evaluations, GPT-6.1 Sol scores 99.7% on ExploitBench at maximum reasoning effort, a result OpenAI cautions may be artificially inflated by potential contamination from exposure to historical vulnerabilities. On ExploitBench Internal Port, an internal evaluation using recently disclosed vulnerabilities, the model reaches a 21.5% arbitrary code-execution success rate, versus 5.5% for GPT-6 Sol and 31.5% for Astra. It scores 78.8% on SEC-Bench Pro and reaches a 35.1% intended-vulnerability success rate per attempt on ExploitGym. OpenAI states that the model’s reported biological results did not cross the indicative Critical thresholds, and that it is taking a phased trusted-access approach for cyber through its Daybreak program, as it did with Astra.

OpenAI reports that GPT-6.1 Sol performs on par with GPT-6 Astra on its HealthBench evaluations, with length-adjusted scores within 0.5 percentage points of Astra across all four evaluations, and scores 57.9 overall on MentalHealthBench versus 58.7 for Astra at maximum reasoning effort.

The launch follows OpenAI’s introduction of GPT-6 Astra on September 3, 2026, and of GPT-6 Sol and GPT-6 Luna on September 22, 2026, when the company cut API prices by 50% from their GPT-5.6 promotional pricing. GPT-6 Sol moved to $2 per million input tokens and $10 per million output tokens, the same standard prices GPT-6.1 Sol now carries, and the earlier release also brought prompt-caching improvements offering 90% discounts on cached input-token reads.

OpenAI says it will offer GPT-6.1 Sol Ultrafast in the coming days, with up to 8x faster token generation than its standard speed in Codex. The DevDay recap describes Ultrafast as a premium speed tier reaching 300 tokens per second in Codex and up to 6x faster token generation in the API, with GPT-6 Astra Ultrafast available now and GPT-6.1 Sol Ultrafast coming soon.