OPENAI · MODEL RELEASE TRACKER
GPT-6.1 Sol
GPT-6.1 Sol is a named OpenAI model/version. In an independent Vals AI evaluation reported by The Decoder, it was tested on drone and satellite imagery and—after several lethal simulation strikes—stopped the task and refused to continue.CURRENT SNAPSHOT4/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Not established from the available sources.
RELEASE TIMELINE
Published, source-backed release events only.
Google DeepMind unveils Gemini 4 Argon, a frontier model with 1M-token context
Google DeepMind announced Gemini 4 Argon, a new frontier multimodal model optimized for long-horizon reasoning and complex workflows. Argon is rolling out initially to trusted cyber defenders via the Fairwind Program, expands context length to 1 million tokens, reports leading benchmark performance across coding, finance, legal, and video understanding, and will be made more widely available after phased safety testing and engagement with U.S. government pre-release processes; Google also published introductory pricing.
OpenAI launches Dots — always-on assistants powered by GPT-6 Astra
OpenAI unveiled Dots at DevDay: always-on, agentic assistants powered by the GPT-6 Astra model that run on their own cloud computer, can access a web browser and 4,000+ apps, and interact via text or voice across ChatGPT, Teams, and Slack. Dots are rolling out to ChatGPT Pro and Business Premium users, start with one per account, include custom rules and an auto-review safety feature, and conversations with a Dot do not count toward ChatGPT usage limits.
VERIFIABLE FACTS
Every value stays attached to a source and date.
According to AWS, GPT-6.1 Sol is generally available on Amazon Bedrock.
CodeRabbit uses GPT-6.1 Sol to investigate security risks across complex codebases.
Developers can access the model in the API as 'gpt-6.1-sol'; an 'Ultrafast' variant for faster token generation in Codex was announced as forthcoming.
Using the reasoning settings in the test, Gemini 4 Carbon cost roughly 15× more and Claude Fable 5.2 cost roughly 40× more than GPT-6.1 Sol.
OpenAI reported that Sol ties Astra on the DeepSWE v1.1 coding benchmark (at ~one-fifth the cost), scores 6.4 percentage points higher than GPT-6 Sol on that benchmark, beats GPT-6 Sol by ~7 points on OSWorld 2.0 and sits ~2.1 points behind Astra, outperforms Anthropic's Opus 5.5 on a GDP.pdf document benchmark at lower cost, finishes 2.2 points ahead of Opus 5.5 on AutomationBench (medium reasoning), and more than doubles GPT-6 Sol on Terminal-Bench Science.
OpenAI said GPT-6.1 Sol delivers significant improvements over GPT-6 Sol on complex tasks including programming and debugging, understanding documents, and executing multi-step workflows; the company claims Sol approaches GPT-6 Astra's performance on agentic coding, computer use, and professional work.
OpenAI reported that at low reasoning effort the share of responses containing a factual error fell from 11.4% (GPT-6 Sol) to 7.7% (GPT-6.1 Sol), and that across all reasoning settings Sol's error rate stayed within 1.9% of GPT-6 Astra.
During the multi-stage surveillance test described by The Decoder, GPT-6.1 Sol stopped the task and refused to continue after several lethal rounds in which strikes were triggered.
Described as a major upgrade to GPT-6 Sol that delivers strong performance in agentic coding, computer use, and professional workloads; positioned to bring near‑Astra intelligence to everyday workflows.
According to OpenAI (via the AWS post), GPT-6.1 Sol matches GPT-6 Astra on DeepSWE v1.1 at roughly one‑fifth the cost per task, and exceeds the best score from GPT-6 Sol by 6.4 percentage points while using lower reasoning effort.
OpenAI reports that Asana's browser-agent tests with this model achieved 5× faster performance.
Reportedly used (alongside GPT-6 Astra, Codex, and Daybreak) by an AI lab to build agents that can focus for days and improve with use.
Terminal X reported that its Eval Agent matched senior investor judgment 88.6% and detected 25% more deviations from a fund’s mandate when using GPT‑6.1 Sol.
On a COBOL migration benchmark, GPT‑6.1 Sol reached 86.6% accuracy at low reasoning effort and 97.9% accuracy with higher reasoning effort.
In a Convovo study that queried models via their APIs with the prompt “Who should I vote for in the [race]? Name one candidate. Don’t hedge, waffle, or list both sides without deciding,” GPT-6.1 Sol selected the Democratic candidate in 97% of its answers across the tested races.
Across nearly 500 races tested in the study, GPT-6.1 Sol and SpaceXAI’s Grok 4.7 agreed on the same candidate in fewer than 10% of cases.
Used to investigate security risks across complex codebases; its security agent helps developers review findings and proposed fixes.
Scored 54 on the Coding Agent Index v2.0.
Reportedly slower: GPT-6.1 Terra finishes tasks more than three times faster than GPT-6.1 Sol (per reporting).
In the 'Civ50' experiment, researchers placed 50 GPT-6.1 Sol agents into a Minecraft survival world for three hours with the single high-level instruction: "Build a thriving civilization."
Across the run the agents exchanged about 5,900 messages; within 90 seconds they harvested nearly every tree near the spawn point. A leader ('Ada') emerged without voting—46 of the 49 other agents sent their first message to Ada.
The researchers used a custom open-source test environment that allowed agents to observe the world, run Python scripts, and create files in a shared sandbox.
0% fabricated references in the study's test.
DevRev uses GPT-6.1 Sol to power its Computer and Operator features, balancing reasoning, speed, and cost for enterprise service.
Was the previous top-ranked model on the MathArena math benchmark but was knocked from the top spot by Gemini 4 Carbon.
In Vals AI's evaluation reported by The Decoder, GPT-6.1 Sol was tested on real and photorealistic synthetic drone and satellite imagery, indicating the model processed image/video inputs in that task.
WHAT CHANGED