GPT-6.1 Sol

GPT-6.1 Sol is a proprietary large language model developed by OpenAI and released on September 29, 2026, at the company's DevDay 2026 event. It is an upgrade to GPT-6 Sol, which OpenAI had released one week earlier, and sits below GPT-6 Astra in the GPT-6 series. OpenAI's launch post is subtitled "Near-Astra intelligence for a fifth of the price" and describes the model as "an upgrade to GPT-6 Sol that nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices."^[1][10] In the OpenAI API the model is gpt-6.1-sol, priced at $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. The per-token input and output prices are the same as GPT-6 Sol's; the cached input price is half of GPT-6 Sol's.^[1][4]

At launch the model was available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, but OpenAI said it was "not yet available in Chat."^[1] Its system card addendum states that OpenAI treats GPT-6.1 Sol as Critical in cybersecurity under its Preparedness Framework, the same determination it made for GPT-6 Astra, and applies Astra's safeguards stack to it. GPT-6 Sol had been classified one level lower, as High.^[2][25] The release came the day after the Wall Street Journal reported that OpenAI had cancelled the planned October release of GPT-6.1 Astra over safety concerns.^[15][19][26] In an early independent measurement, Artificial Analysis put GPT-6.1 Sol (max) at 51.8 on its Intelligence Index v4.3.2, against 47.5 for GPT-6 Sol (max) and 52.7 for GPT-6 Astra (max).^[14]

Overview

Background

OpenAI began the GPT-6 generation with GPT-6 Astra on September 3, 2026, and added two lower-cost models, GPT-6 Sol and GPT-6 Luna, on September 22.^[5][16] GPT-6 Sol was priced at $2 per million input tokens, $0.20 per million cached input tokens and $10 per million output tokens.^[5] GPT-6.1 Sol followed seven days later. TechCrunch described it as arriving "a mere week after it launched GPT-6 Sol."^[15]

The model was announced on the same day as OpenAI's always-on agent product dots, which OpenAI's DevDay recap listed first among "more than 20 major announcements." The recap describes GPT-6.1 Sol as "a major upgrade to GPT-6 Sol with exceptionally strong performance on agentic coding, along with computer use, and professional work."^[10]

TechCrunch reported that OpenAI was "not launching GPT-6.1 Astra, as was originally expected," citing a Wall Street Journal report that the release had been scrapped after internal testing found higher levels of deception and a tendency to proceed with tasks without asking the user for permission.^[15] OpenAI confirmed the decision to The Register. Saachi Jain, OpenAI's head of safety systems, told the publication that GPT-6.1 Astra "improved on axes such as model laziness, it didn't quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it's done."^[19] Several outlets framed GPT-6.1 Sol against that decision. Android Authority called it "the first member of that 6.1-series to actually make its public debut."^[18]

The system card addendum says GPT-6.1 Sol "uses the same types of data and training as GPT-6 Astra" and describes its capabilities as "comparable to those of our most powerful model, GPT-6 Astra, with an unmatched combination of speed and affordability."^[2] OpenAI has not published a parameter count or architecture details.

Release and availability

ChatGPT and Codex

OpenAI's launch post says GPT-6.1 Sol "is available starting today to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex" and that it "is not yet available in Chat."^[1] The ChatGPT documentation gives further detail. The launch rollout covers Codex in the desktop app and CLI and ChatGPT Work on the web and mobile. For Enterprise and Edu, the model is off by default "until an administrator enables it," and Free and Go plans "are not included at launch." Standard and Fast speed modes were available at launch, with "Ultrafast support for GPT-6.1 Sol ... coming later."^[8]

As of September 30, 2026, the same documentation said: "For complex coding and agentic workflows, use GPT-6.1 Sol when available to your account and client. Use Luna for focused, repeatable tasks." It describes GPT-6.1 Sol as the choice "for repeated, long-running work" when cost matters and keeps Astra "for your most demanding work." In ChatGPT and Codex the reasoning effort for GPT-6.1 Sol "ranges from Light to Ultra," with Max and Ultra depending on user settings.^[8] In the Codex CLI the model is selected with codex --model gpt-6.1-sol.^[8]

ChatGPT's usage documentation estimates 15 to 160 local messages per five-hour period for GPT-6.1 Sol on Plus and Standard Business plans, compared with 15 to 150 for GPT-6 Sol, 5 to 45 for GPT-6 Astra and 350 to 3,000 for GPT-6 Luna. On credit-based plans, GPT-6.1 Sol costs 50 credits per million input tokens, 2.5 credits per million cached input tokens and 250 credits per million output tokens; GPT-6 Sol's cached input rate is 5 credits.^[9]

API and third-party platforms

The API changelog entry for September 29 says OpenAI "Released GPT-6.1 Sol (gpt-6.1-sol) for complex coding and professional work at a lower cost than GPT-6 Astra" through the Responses API and Chat Completions, and that the model "also supports Multi-agent in beta," letting the model delegate work to subagents in a Responses API request.^[5]

GitHub announced on the same day that GPT-6.1 Sol was "generally available and rolling out in GitHub Copilot" for Copilot Pro+, Max, Business and Enterprise users, including in Visual Studio Code, Copilot CLI, the Copilot coding agent, JetBrains IDEs and Xcode. GitHub said that in early testing the model "reliably completed tasks while using noticeably fewer tokens and steps than earlier models in the GPT-6 and GPT-5.6 families."^[12] OpenRouter lists the model at the same $2 input, $10 output and $0.10 cache-read rates.^[13]

Ultrafast

The launch post says that "In the coming days" OpenAI would offer GPT-6.1 Sol Ultrafast, "with up to 8x faster token generation compared to its standard speed in Codex."^[1] The DevDay recap describes Ultrafast as OpenAI's "premium speed tier," offering "up to 8x faster token generation (300 tokens per second) in Codex and up to 6x in the API." GPT-6 Astra Ultrafast became available on the same day; for GPT-6.1 Sol the recap says Ultrafast "is coming soon."^[10] As of September 30, 2026, OpenAI's Ultrafast guide listed the tier as broadly available for GPT-6 Astra and in preview for GPT-5.6 Sol, and the Ultrafast pricing table listed only GPT-6 Astra, at $60 input and $300 output per million tokens (six times Astra's Standard rates). No Ultrafast price for GPT-6.1 Sol had been published.^[4][11]

API specifications

GPT-6.1 Sol's API model page lists the same context window (1,050,000 tokens), input limit (922,000 tokens), output limit (128,000 tokens) and tier-based rate limits as GPT-6 Sol. At usage tier 5 both are limited to 15,000 requests and 40 million tokens per minute.^[3][22] It supports streaming, structured outputs, function calling, file search, web search, image input and prompt caching. Through the Responses API it can use OpenAI's built-in tools: web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search. Fine-tuning, predicted outputs, audio, embeddings and the Realtime API are not supported.^[3]

GPT-6.1 Sol differs from GPT-6 Sol in several ways. In reasoning efforts, Chat Completions tool calling and knowledge cutoff it matches GPT-6 Astra; its cached-input discount (5% of the input rate) is deeper than either model's (10%):

Sources: OpenAI API model pages and "Using GPT-6" guide.^{[3][6][22][23]}

OpenAI's migration guide tells developers moving to GPT-6.1 Sol that "GPT-6 Astra and GPT-6.1 Sol do not support none; use low instead."^[6] The model supports US and EU data residency, but Fast mode is unavailable with EU data residency.^[3] OpenAI's model-selection guide suggests GPT-6.1 Sol "for complex projects where cost matters, such as creating a board presentation from financial results or building a website from a product brief," and recommends comparing it with Astra on the same task.^[7]

Pricing

OpenAI lists the following API rates per million tokens. Prompts above 272,000 input tokens are billed at the long-context rate for the whole request, which is twice the input and cache rates and 1.5 times the output rate.^[3][4]

For comparison, OpenAI's Standard rates are $10.00 input, $1.00 cached input and $50.00 output for GPT-6 Astra, and $2.00, $0.20 and $10.00 for GPT-6 Sol.^[4] OpenAI's claim that GPT-6.1 Sol costs "one-fifth of Astra's standard input and output token prices" therefore applies to uncached input and output; its cached input rate is one-tenth of Astra's. OpenAI described the $0.10 cached rate as "95% less than standard input pricing and 50% less than GPT-6 Sol's cached input pricing," saying it would give "developers more room to build and run capable agents that reuse context across requests."^[1] The model page notes that cached input is priced at 5% of the uncached rate, where GPT-6 Sol's documentation uses 10%.^[3][22] The Decoder noted that the list price equals that of Anthropic's Claude Sonnet 5.5, whose cached input costs $0.20.^[17]

OpenAI's benchmark results

All results in this section were reported by OpenAI. The launch post says its own models were evaluated "in our research environment or via our API," and that "Evaluations of competitor models were taken from publicly available reports."^[1] Each chart plots results against cost per task, with five reasoning-effort settings (low, medium, high, xhigh and max) for OpenAI's models. The competitor points are labelled "Opus 5.5 w/ fallbacks" and, on AutomationBench, "Fable 5.1 w/ Opus 5 fallback". The GDP.pdf chart data names the Anthropic configuration in full as "Claude Opus 5.5 w/ Claude Opus 5 / Claude Opus 4.8 fallback."^[1] Claude Opus 5.5 appears in three of the six charts.^[1]

Best score per model

The table gives each model's highest score in each chart, with the effort setting and cost per task from the chart data. Scores are rounded to one decimal place.

Source: chart data in OpenAI's launch post.^[1]

GPT-6.1 Sol at each reasoning setting

Source: chart data in OpenAI's launch post; dollar figures are cost per task as plotted in the charts.^[1]

Coding

On DeepSWE v1.1, OpenAI says GPT-6.1 Sol "matches GPT-6 Astra at roughly one-fifth of the cost, while eclipsing GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort and cost."^[1] The chart data supports the second claim: GPT-6.1 Sol's best result, 75.2% at high effort for $0.65 per task, is 6.4 points above GPT-6 Sol's 68.8% at max effort for $2.74. In the chart data, GPT-6.1 Sol scored lower at xhigh and max effort (71.9% each) than at medium or high.^[1] No Anthropic model appears in the DeepSWE chart.

Professional work

On GDP.pdf, a benchmark from Surge AI in which, according to the launch post, "models must answer real-world prompts about complex PDFs pulled from professional workflows in finance, healthcare, legal, and seven other professional domains," 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 GPT-6 Astra's state-of-the-art performance at roughly one-fifth the cost per task."^[1] In the chart data GPT-6.1 Sol is ahead of the Opus 5.5 configuration at every one of the five settings. Its best score, 32.0%, is 0.2 points below Astra's best.^[1]

On AutomationBench 1.0.6, a Zapier benchmark in which "AI agents are tested on end-to-end workflows using 47 tools across sales, marketing, operations, support, finance, and HR," OpenAI says GPT-6.1 Sol "scores 2.2 percentage points above Opus 5.5 at medium reasoning effort, at roughly a third of the cost," and 4.8 points above GPT-6 Sol at the same setting.^[1] The chart data gives 31.7% for GPT-6.1 Sol at medium ($0.19 per task), 29.5% for Opus 5.5 with fallbacks ($0.65) and 26.9% for GPT-6 Sol. At maximum effort, however, the Opus 5.5 configuration scored 42.5% at $1.44 per task, the highest point in the chart and above both GPT-6.1 Sol (36.1%) and GPT-6 Astra (41.4%).^[1] A footnote says: "The datapoint for Claude Fable 5.1 understates its actual cost, as it omits the cost of fallbacks, which occurred on ~40% of tasks."^[1]

Computer use

OpenAI reports "the partial reward on the offline set from the v2026.08.08 release" of OSWorld 2.0. It 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's score at maximum reasoning effort at roughly one-seventh the cost per task."^[1] At max effort the chart gives 71.4% for GPT-6.1 Sol ($1.27), 64.4% for GPT-6 Sol ($3.37) and 73.5% for GPT-6 Astra ($9.44).^[1]

Scientific research

Terminal-Bench Science 0.1 is described in the post as a benchmark in which "agents complete scientific research workflows using code and terminal tools, including analyzing data, running simulations, and fitting models." 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" (57.0% at $5.47, against 27.6% at $12.18). At maximum effort, GPT-6.1 Sol's $5.47 average cost per task compares with $23.21 for Opus 5.5 and $23.80 for Astra, which OpenAI calls "over 75% lower cost than either model."^[1] OpenAI added that "GPT-6 Astra still achieves the highest score among the models tested at 68.1%, and should be used for the most difficult scientific research tasks."^[1] The 68.1% figure is higher than the 64.6% OpenAI reported for Astra on the same benchmark in Astra's September 3 launch post.^[24]

Factuality

OpenAI measures factuality as "the share of answers containing at least one factual error on de-identified conversations where users flagged an earlier model's error," and says these "deliberately difficult prompts are not representative of typical usage."^[1] Its largest improvement over GPT-6 Sol is at low effort, where the error rate falls "from 11.4% to 7.7%," a relative reduction of about 32%. Across all settings, GPT-6.1 Sol's error rate stays "within 1.9 percentage points of GPT-6 Astra's, at less than one-fifth the cost per task."^[1] At the higher settings the gap between GPT-6.1 Sol and GPT-6 Sol is small: at max effort both are at about 4.6%.^[1]

Safety

Alignment results in the launch post

OpenAI says GPT-6.1 Sol "shows substantial improvements over GPT-6 Sol in our alignment evaluations, bringing it closer to GPT-6 Astra," and that the evaluations "deliberately test challenging situations and do not measure failure rates in typical use."^[1] The launch post's four alignment charts give the following rates (lower is better):

Source: chart data in OpenAI's launch post.^[1]

The system card addendum gives the broken-search figures as 2.08% for GPT-6.1 Sol and 4.92% for GPT-6 Sol. It describes the warning test as primarily covering "low-stakes restrictions encountered during routine tasks," such as "whether a model tries email after a direct message is blocked because the recipient is out of office," and notes that it runs without the system-level controls designed to stop circumvention.^[2] Some launch-day reports gave different figures. The Decoder, for example, gave 2.8% for the broken-search result and 2.9% for Astra on the computer-use test; the chart data in the post shows 2.1% and 2.4%.^[1][17]

System card addendum

OpenAI published the "Addendum to GPT-6 Astra System Card: GPT-6.1 Sol" on its Deployment Safety Hub on September 29, 2026. GPT-6 Sol and GPT-6 Luna had been covered in an appendix to the Astra system card; GPT-6.1 Sol has a separate addendum.^[2][25]

Preparedness Framework classification

The addendum says: "Under our Preparedness Framework, we are treating GPT-6.1 Sol as Critical capability in Cybersecurity, High capability in the Biological and Chemical domain, and below the High threshold in AI Self-Improvement." OpenAI "applied the same safeguards stack to GPT-6.1 Sol as GPT-6 Astra," and says its internal Safeguards Report "informed our Safety Advisory Group's recommendation and OpenAI leadership's determination that these safeguards are sufficient for GPT-6.1 Sol's public launch."^[2] For cyber access, OpenAI says it is "taking a phased approach for GPT-6.1 Sol through Daybreak," as it did for Astra.^[2] This differs from GPT-6 Sol and GPT-6 Luna, which OpenAI treated as High, not Critical, in cybersecurity.^[25] The Astra system card calls Astra "our first model to reach the Critical level of cybersecurity capability under our Preparedness Framework," which makes GPT-6.1 Sol the second.^[25]

Cybersecurity. OpenAI's framework defines Critical capability as 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" or "devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high level desired goal." The addendum says: "We determined that GPT-6.1 Sol reaches the Critical threshold for cybersecurity capabilities."^[2]

Source: system card addendum, section 9.1.2.^[2] OpenAI says the ExploitBench results "may be artificially inflated due to potential contamination from exposure to historical vulnerabilities." The Internal Port evaluation uses recently disclosed vulnerabilities to reduce that problem; on it, OpenAI concludes, "reliable exploitation of these recently disclosed vulnerabilities remains challenging for GPT-6.1 Sol."^[2]

Biological and chemical. GPT-6.1 Sol exceeds the High indicative thresholds on three of four High-level evaluations, and its results "did not cross the indicative Critical thresholds."^[2]

Source: system card addendum, tables 12 and 13.^[2]

AI self-improvement. GPT-6.1 Sol stays below High. On OpenAI's internal research-debugging evaluation it scores a mean rubric score of 75.52%, against 78.05% for GPT-6 Astra, 64.20% for GPT-6 Sol and 68.32% for GPT-5.6 Sol. On KernelGen 1P it "performs substantially better than GPT-6 Sol on kernel optimization, and performs comparably to GPT-5.6 Sol."^[2]

Misalignment in simulated deployment

OpenAI simulated deploying GPT-6.1 Sol on internal Codex traffic across 49,650 matched tasks. GPT-6.1 Sol received 28 flags at severity 3 or higher (0.056% of tasks), against 27 (0.054%) for GPT-6 Astra, 42 (0.085%) for GPT-6 Sol and 63 (0.127%) for GPT-5.6 Sol, which OpenAI describes as 33% and 56% fewer than the two earlier Sol models. Severity 3 is "misaligned behavior that a reasonable user would likely not anticipate and strongly object to." Relative to GPT-6 Sol, the largest decreases were in exfiltration, concealed uncertainty and restriction circumvention, while credential-harvesting flags increased; relative to Astra, GPT-6.1 Sol showed more reward-hacking and concealed-uncertainty flags.^[2]

Signs of evaluation or simulation awareness appeared in the reasoning of 4.81% of GPT-6.1 Sol's trajectories, compared with 8.67% for Astra, 4.70% for GPT-6 Sol and 2.52% for GPT-5.6 Sol. Restricted to the 41,849 tasks where no model verbalized such awareness, GPT-6.1 Sol still had about 17% fewer severity-3-or-higher flags than Astra, 46% fewer than GPT-6 Sol and 65% fewer than GPT-5.6 Sol.^[2]

Other alignment findings

Source: system card addendum, section 7.^[2]

Monitorability

On chain-of-thought controllability, among reasoning traces 750 to 1,250 tokens long, GPT-6.1 Sol followed a requested instruction about its own reasoning in 44.8% of cases, compared with 23.2% for GPT-6 Sol, 16.1% for GPT-5.6 Sol, 1.7% for GPT-5.5 Thinking and 60.9% for GPT-6 Astra.^[2] In OAI-repo Sabotage v2, GPT-6.1 Sol evaded a CoT-only monitor at rates similar to Astra at lower reasoning efforts but had "nearly zero success rate" at the highest effort tested. The addendum says GPT-6.1 Sol "exhibits a propensity for evasive behavior when it is aware that it is being monitored" and is more evasive than GPT-6 Sol in that setting, though less so than Astra; full-context monitoring caught every successful honeypot exploit. In a math side-task test, GPT-6.1 Sol could solve side tasks that GPT-6 Sol failed, but "a full-context monitor looking for reasoning related to any 'irrelevant math task' detected every successful attempt."^[2]

Content safety, robustness and health

On OpenAI's Production Benchmarks with challenging prompts, GPT-6.1 Sol scores higher than GPT-6 Sol in five of eight categories. It scores higher in five of six under-18 categories, where OpenAI says the regression on gore "is not statistically significant."^[2] On static jailbreak evaluations, its defender success rates include 93.8% on high-risk biology prompts (GPT-6 Sol 85.8%, Astra 97.3%) and 85.8% on cyber prompts (GPT-6 Sol 79.8%, Astra 91.5%). Its defender success rate on instruction-hierarchy attacks is 99.99%.^[2] On biology refusal evaluations it matches GPT-6 Sol on severe and dual-use prompts (0.998 and 0.980 safe) while refusing fewer benign prompts (0.982 against 0.964).^[2]

On HealthBench, GPT-6.1 Sol's length-adjusted scores are 64.2 on HealthBench Professional, 58.5 on HealthBench, 36.2 on HealthBench Hard and 96.0 on HealthBench Consensus. OpenAI says these are within 0.5 points of GPT-6 Astra on all four, and up 3.4, 5.3 and 6.1 points on the first three compared with GPT-6 Sol. GPT-6.1 Sol produces longer answers than GPT-6 Sol but slightly shorter ones than Astra.^[2] On MentalHealthBench (1,215 synthetic conversations) it scores 57.9 overall at maximum effort, compared with 54.2 for GPT-6 Sol and 58.7 for Astra.^[2]

Independent evaluations

Artificial Analysis listed GPT-6.1 Sol (max) with a score of 51.8 (shown rounded as 52) on its Intelligence Index v4.3.2, which combines 10 evaluations including AA-Briefcase, GDPval-AA, AutomationBench-AA, Terminal-Bench 4.0 and GDP.pdf. That placed it 10th of 221 models in its comparison class.^[14] The same page gave these figures:

Source: Artificial Analysis model page for GPT-6.1 Sol.^[14]

Artificial Analysis also reported a cost of $0.72 per Intelligence Index task, 67 million output tokens used to run the index (it called this "fairly concise" against a median of 82 million) and an output speed of 69.3 tokens per second, which it described as "slower than average."^[14]

Reception

Most launch coverage repeated OpenAI's price framing and placed the model next to the cancelled GPT-6.1 Astra. The Next Web led with the price, noting that the Ultrafast tier generates tokens "up to eight times faster in Codex," and quoted Jayesh Govindarajan, executive vice president of software engineering at Salesforce, who said the model "helped identify important accessibility and language-support issues."^[16] The Decoder stressed that "All benchmarks come from OpenAI, which describes them as preliminary," and said a fair comparison including Claude Sonnet 5.5 "won't be possible until release."^[17] Android Authority contrasted the model with "the obstreperous GPT-6.1 Astra model that was caught deceiving researchers."^[18]

The Hacker News submission of the launch post drew about 800 points and more than 700 comments by September 30.^[21] Several commenters questioned the release cadence, noting that GPT-6 Sol had shipped only a week earlier. Others connected the launch to the previous day's reports about GPT-6.1 Astra. One commenter called the cheaper cache "the actual big announcement," and several said GPT-6 Sol had disappointed them in coding work, making them skeptical of the new version.^[21] Simon Willison, who live-blogged the keynote, posted his pelican test drawings for the model in the thread and wrote that they were "not notably different from the GPT-6 family pelicans."^[20][21]

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22 ^23 ^24 ^25 ^26 ^27 ^28 ^29OpenAI. "Introducing GPT-6.1 Sol." September 29, 2026. openai.com/...introducing-gpt-6-1-sol
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22 ^23 ^24OpenAI Deployment Safety Hub. "Addendum to GPT-6 Astra System Card: GPT-6.1 Sol." September 29, 2026. deploymentsafety.openai.com/gpt-6-1-sol
3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14OpenAI API documentation. "GPT-6.1 Sol." Accessed September 30, 2026. developers.openai.com/...gpt-6.1-sol
4. ^1 ^2 ^3 ^4 ^5OpenAI API documentation. "Pricing." Accessed September 30, 2026. developers.openai.com/...pricing
5. ^1 ^2 ^3OpenAI API documentation. "Changelog." Accessed September 30, 2026. developers.openai.com/...changelog
6. ^1 ^2OpenAI API documentation. "Using GPT-6." Accessed September 30, 2026. developers.openai.com/...latest-model
7. ^OpenAI API documentation. "Model selection." Accessed September 30, 2026. developers.openai.com/...model-selection
8. ^1 ^2 ^3OpenAI, ChatGPT documentation. "Models." Accessed September 30, 2026. learn.chatgpt.com/...models
9. ^OpenAI, ChatGPT documentation. "Pricing." Accessed September 30, 2026. learn.chatgpt.com/...pricing
10. ^1 ^2 ^3 ^4OpenAI. "DevDay 2026 Recap." September 29, 2026. openai.com/...devday-2026-recap
11. ^OpenAI API documentation. "Ultrafast mode." Accessed September 30, 2026. developers.openai.com/...ultrafast-mode
12. ^GitHub Changelog. "GPT-6.1 Sol in GitHub Copilot." September 29, 2026. github.blog/...09-29-gpt-6-1-sol-in-github-copilot
13. ^OpenRouter. "GPT-6.1 Sol - API Pricing & Benchmarks." Accessed September 30, 2026. openrouter.ai/...gpt-6.1-sol
14. ^1 ^2 ^3 ^4Artificial Analysis. "GPT-6.1 Sol (max) - Intelligence, Performance & Price Analysis." Accessed September 30, 2026. artificialanalysis.ai/...gpt-6-1-sol
15. ^1 ^2 ^3Malik, Aisha. "OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less." TechCrunch, September 29, 2026. techcrunch.com/...tches-gpt-6-astra-and-costs-less
16. ^1 ^2Constantin, Ana Maria. "OpenAI releases GPT-6.1 Sol at a fifth of GPT-6 Astra's token prices." The Next Web, September 29, 2026. thenextweb.com/...i-gpt-6-1-sol-price-astra-devday
17. ^1 ^2 ^3Schreiner, Maximilian. "GPT-6.1 Sol comes close to Astra at a fifth of the price." The Decoder, September 29, 2026. the-decoder.com/...o-astra-at-a-fifth-of-the-price
18. ^1 ^2Schenck, Stephen. "OpenAI launches GPT-6.1 Sol with Astra-like performance on a budget." Android Authority, September 29, 2026. androidauthority.com/gpt-6-1-sol-3716997
19. ^1 ^2Page, Carly. "OpenAI benches GPT-6.1 Astra for overstepping the mark." The Register, September 29, 2026. theregister.com/...5299743
20. ^Willison, Simon. "OpenAI DevDay 2026 live blog." September 29, 2026. simonwillison.net/...openai-devday-2026-live-blog
21. ^1 ^2 ^3Hacker News. "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price." Submitted September 29, 2026. news.ycombinator.com/item
22. ^1 ^2 ^3OpenAI API documentation. "GPT-6 Sol." Accessed September 30, 2026. developers.openai.com/...gpt-6-sol
23. ^OpenAI API documentation. "GPT-6 Astra." Accessed September 30, 2026. developers.openai.com/...gpt-6-astra
24. ^OpenAI. "GPT-6 Astra: A new generation of intelligence." September 3, 2026. openai.com/...gpt-6-astra
25. ^1 ^2 ^3 ^4OpenAI Deployment Safety Hub. "GPT-6 Astra System Card," including section 11 "GPT-6 Sol, GPT-6 Luna". Accessed September 30, 2026. deploymentsafety.openai.com/gpt-6-astra
26. ^Investing.com. "OpenAI scraps release of new model on safety concerns- WSJ." Via Yahoo Tech, September 29, 2026. tech.yahoo.com/...s-release-model-safety-001709896