OpenAI’s new GPT-6.1 Sol undercuts its own Astra flagship
OpenAI on Tuesday launched GPT-6.1 Sol, the newest version of its workhorse model, only a week after launching GPT-6 Sol. The post OpenAI’s new GPT-6.1 Sol undercuts its own Astra flagship appeared first on The New Stack .
OpenAI unveiled GPT-6.1 Sol, an enhanced version of its GPT-6 Sol model, just a week after releasing the initial GPT-6 Sol. The company touts the new model as nearly matching GPT-6 Astra's intelligence for coding, computer use, and professional tasks, but at a fraction of Astra's token prices. The pricing for GPT-6.1 Sol remains the same as before, at $2 per million input tokens and $10 per million output tokens, with a discounted rate of $0.10 per million tokens for cached input.
GPT-6.1 Sol is now accessible via the API and to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. An Ultrafast version of the model is slated for Codex, boasting up to 8x faster token generation than the standard speed. Despite being only a week old, GPT-6.1 Sol demonstrates significant improvements, ranking similarly to the expensive GPT-6 Astra in benchmarks while costing considerably less.
In coding benchmarks, GPT-6.1 Sol outperforms GPT-6 Sol by 6.4 percentage points on DeepSWE 1.1, reaching results comparable to GPT-6 Astra at just one-fifth the cost. On the GDP.pdf benchmark, which tests the models' ability to answer questions about complex PDF documents, GPT-6.1 Sol scored around 32% compared to 29% for Opus 5.5, a similar performance to GPT-6 Astra at one-fifth the cost per task.
GPT-6.1 Sol excels in computer use, outperforming its predecessor by seven percentage points at maximum reasoning and half the cost, matching Astra's performance.
GPT-6.1 Sol also exhibits fewer factual errors, with the rate falling from 11.4% with GPT-6 Sol to 7.7%—a 32% reduction. This improvement is observed in deliberately difficult conversations, not representative of typical use. OpenAI reports that GPT-6.1 Sol is better at respecting user intent and safety constraints, disclosing broken search tools only 2.8% of the time, whereas Astra's guessing rate is higher.
Agents based on GPT-6.1 Sol also never attempt to circumvent automated safety reviewers' decisions to block their actions.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
