A week after launching GPT-6 Sol and Luna, OpenAI went back to the model family with a mid-tier upgrade. Announced on stage at DevDay 2026 on Tuesday at Fort Mason in San Francisco, GPT-6.1 Sol is pitched squarely at developers running agents: near-flagship intelligence on coding, computer use, and professional work, with pricing that undercuts the top of the line by a factor of five.

The numbers are the headline. Developers can call it in the API as gpt-6.1-sol at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens — exactly one-fifth of GPT-6 Astra's $10/$50 standard rates, per the model card in OpenAI's announcement. Cached input is 95% below its own standard input price and half of GPT-6 Sol's cached-input rate. The model carries a 1-million-token context window with up to 128,000 output tokens, per Neowin's DevDay coverage.

OpenAI's benchmarks — with the usual caveat#

OpenAI's launch materials lean hard on cost-per-task rather than raw capability. On DeepSWE v1.1, a benchmark for complex software-engineering work in real codebases, the company says GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth of the cost, while beating GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort. On GDP.pdf — answering professional questions from complex PDF documents — it scores above Claude Opus 5.5 (with fallbacks) at less than half the cost per task.

Editorial illustration: a small efficient robot sprinting ahead of a giant expensive robot on a glowing track beside a programmer's desk
AI-generated illustration for this article.

The agent-focused numbers go further. On AutomationBench, which tests multi-step business workflows, OpenAI reports GPT-6.1 Sol scores 2.2 percentage points above Opus 5.5 at medium reasoning effort and at roughly a third of the cost. On the OSWorld 2.0 computer-use set, it improved 7% over GPT-6 Sol at maximum reasoning effort and lands within 2.1 percentage points of Astra at roughly one-seventh of the cost per task. On Terminal-Bench Science 0.1 — scientific workflows spanning data analysis, simulation, and theorem proving — it more than doubles GPT-6 Sol's score, averaging $5.47 per task against $23.21 for Opus 5.5 and $23.80 for Astra, though Astra still took the highest score at 68.1%.

Factuality got its own line: the share of responses containing a factual error fell from 11.4% with GPT-6 Sol to 7.7% with GPT-6.1 Sol at low reasoning effort — about a 32% reduction — staying within 1.9 percentage points of Astra. Two caveats from the fine print: these are OpenAI-run evaluations, and the factuality test used deliberately difficult prompts flagged by users, which the company says aren't representative of typical usage.

Where it ships#

GPT-6.1 Sol is available from September 29 to all Plus, Pro, Business, Enterprise, and Edu users inside ChatGPT Work and Codex, and to developers through the API — though not yet in Chat. A speed tier called GPT-6.1 Sol Ultrafast, offering up to eight times faster generation in Codex, is slated to follow in the coming days.

Editorial illustration: an AI agent's cursor moving between layered translucent software windows connected by circuit traces
AI-generated illustration for this article.

The launch lands in a loaded week for OpenAI. It arrived one day after the company delayed its GPT-6.1 Astra update over safety concerns, and it shared the DevDay stage with more than 20 other announcements — including the always-on dots agents and a $500-a-month Pro 500 tier. The message underneath all of it: flagship intelligence is no longer the product — cheap, fast, near-flagship intelligence is.

Sources