We use cookies to enhance your browsing experience, analyze site traffic, and personalize content. By clicking "Accept All", you consent to our use of cookies. You can customize your preferences or reject non-essential cookies.
Learn more about our cookie policyModels · OpenAI · GPT-6.1
Near-Astra capability at a fraction of the price: stronger coding, document analysis and professional workflows, with 1.05M context and ZDR-labelled routing.
OpenAI introduced GPT-6.1 Sol on September 29, 2026 as an upgrade to GPT-6 Sol. It brings coding, computer-use and professional-work performance closer to GPT-6 Astra without its token price. The regular model is now listed on AI·Collab as OpenAI: GPT-6.1 Sol.
Try it for real-codebase debugging, complex document analysis and well-defined multi-step work where both quality and cost matter.
OpenAI recommends Astra for the hardest scientific research. Near-Astra does not mean identical performance on every task.
The story is capability per task cost, not a promise that every answer is faster. OpenAI reports gains across software engineering, complex PDFs and business workflows.
DeepSWE v1.1 evaluates long-horizon coding. Sol 6.1 reaches 75.2% at High effort, above GPT-6 Sol's best 68.8% at Max, at approximately 76% lower task cost in OpenAI's testing.
On GDP.pdf, OpenAI reports performance above Opus 5.5 with fallbacks at less than half the task cost across tested settings. Documents include tables, charts, diagrams and fine print.
AutomationBench uses 47 tools across sales, finance, operations, support and other functions. Sol 6.1 scores 31.7% at Medium effort.
On deliberately difficult prompts at Low effort, answers containing an error drop from 11.4% to 7.7% versus GPT-6 Sol. This is not the error rate for typical usage; verify important claims.
Effort settings, tools and benchmark versions matter. These results are OpenAI's evaluations, not AI·Collab measurements.
A 6.4-percentage-point gain over GPT-6 Sol's best 68.8%, comparing High versus Max effort. OpenAI reports matching Astra at roughly one-fifth of the task cost.
At Medium effort: +4.8 points over GPT-6 Sol and +2.2 over Opus 5.5. OpenAI notes that its Fable 5.1 cost datapoint excludes fallbacks on about 40% of tasks.
Partial reward on release v2026.08.08 at Max effort versus Astra's 73.5%, at roughly one-seventh of Astra's task cost. Do not directly compare with Sonnet 5.5's OSWorld 2.1 result.
Terminal-Bench Science 0.1 average cost at Max: $5.47 versus $23.21 for Opus 5.5 and $23.80 for Astra. Sol more than doubles GPT-6 Sol's score; Astra leads at 68.1%. These are benchmark costs, not AI·Collab quotes.
Official charts and methodology are linked rather than reproduced here: OpenAI — Introducing GPT-6.1 Sol
Verified October 2, 2026: the live catalogue lists 2,428 input and 12,144 output credits per million tokens for the regular model. These are AI·Collab rates, distinct from OpenAI's API prices.
Input / output base rates. The live catalogue is authoritative if pricing changes.
Per million input / output tokens, unchanged from GPT-6 Sol. Cached input is $0.10, down 50%; cache writes are $2.50.
OpenAI's model card specifies 2× input/cache and 1.5× output rates for the full request above 272K input tokens. The AI·Collab catalogue exposes base credit rates, not a verified long-context billing breakdown.
Confirmed in AI·Collab's catalogue. OpenAI specifies 128K maximum output tokens; provider limits, overhead and application settings still apply.
The regular catalogue entry is marked ZDR. That retention label does not establish EU-only inference. Check current model labels and your organization's policy before sensitive work.
Provider routing and retention metadata can change. Use the catalogue and AI·Collab's ZDR guidance as the current reference.
Computer-use benchmark results do not mean AI·Collab chat controls your desktop. Native tool calling and organization assignment were not independently verified in this launch-content task.
Search for OpenAI: GPT-6.1 Sol. Dropdown access depends on your model library and organization curation; this announcement does not change those policies.
If your organization already curates Sol 6.1, select it directly. No manual setup is needed in that case.
Otherwise use Account Settings → My Custom Models if your account policy permits adding it.
Open chat.aicollab.app and provide a scoped debugging or document-analysis task with clear requirements and source material.
Compare with Sonnet 5.5 or your current model on correctness, revisions and actual credit consumption. Review important conclusions before acting.
OpenAI's September 29, 2026 upgrade to GPT-6 Sol, positioned near Astra for coding, computer use and professional work at one-fifth of Astra's standard input/output token prices.
Verified October 2: 2,428 input and 12,144 output credits per million tokens. Check live rates. OpenAI has separate long-context multipliers above 272K input; the catalogue base rates alone do not establish that billing behavior.
Catalogue availability does not guarantee assignment. If already curated, select it directly; otherwise add it through My Custom Models when your policy permits.
The regular model is marked ZDR. EU data residency is separate; this post does not claim EU-only inference.
Both have the same catalogue base credit rates at publication. Sol 6.1 is a candidate for complex documents, coding and workflows; Sonnet 5.5 emphasizes fast, scoped work. Compare your own tasks, not scores from different benchmark versions.
No such availability is established here. OpenAI's product modes and benchmark tools are not automatically AI·Collab features.
Published October 2, 2026. Benchmark figures are reported by OpenAI and not independently verified by AI·Collab. Task-cost savings depend on workload, tools and effort and are not guaranteed AI·Collab savings. Live catalogue metadata was checked on publication day.
Google's new sparse MoE model: evidence-led legal and finance drafts, coding, multimodal understanding, balanced benchmarks, PAID/ZDR and practical limits.
Read moreAnthropic's fastest Sonnet yet: 70.6% Terminal-Bench 4.0, near-Opus knowledge work, 1M context, ZDR and native agentic tools at the same $2/$10 token price.
Read moreOpenAI's balanced GPT-5.6 model supersedes GPT-5.5 for every user at roughly half the credit price, with Sol and Luna available via My Custom Models.
Read moreCredits explained: real cost data for GPT-5.4, Claude Opus, Gemini 3.1 Pro, Perplexity Sonar Pro and more. See how far 3,000 or 15,000 credits take you.
Read moreGet started today. Access models from OpenAI, Google, Anthropic, Grok and more.
GDPR compliant · Zero data retention · Cancel anytime