GPT-6.1 Sol

What is disclosed

OpenAI released GPT-6.1 Sol on 29 September 2026, a week after GPT-6 Sol.¹¹1OpenAI, API changelog, entries of 22 and 29 September 2026. https://developers.openai.com/api/docs/changelog. Arithmetic: 22 September + 7 days = 29 September. The older model's page now sends readers to the new one.²²2OpenAI, GPT-6 Sol model page, read 29 September 2026. https://developers.openai.com/api/docs/models/gpt-6-sol

It takes text and images and returns text. Audio and video are not supported.³³3OpenAI, GPT-6.1 Sol model page, read 29 September 2026. https://developers.openai.com/api/docs/models/gpt-6.1-sol The window is the same as on GPT-6 Astra, 1,050,000 tokens, and so is the knowledge cutoff, 30 April 2026. A response can run to 128,000 tokens.³³

A system card addendum says it uses "the same types of data and training as GPT-6 Astra".⁴⁴4OpenAI, Addendum to GPT-6 Astra System Card: GPT-6.1 Sol, 29 September 2026, sections on introduction, model data and training, and HealthBench (Table 7). https://deploymentsafety.openai.com/gpt-6-1-sol Neither model has a published size or architecture.

What is new

Against GPT-6 Sol, cached input is half the price, $0.10 instead of $0.20 per million tokens.⁵⁵5OpenAI, API pricing, Standard table, read 29 September 2026. https://developers.openai.com/api/docs/pricing. Arithmetic: 0.10 / 0.20 = 0.5; 2 x 2 = 4; 10 x 1.5 = 15; against Astra, 2 / 10 = 0.2 and 10 / 50 = 0.2. The knowledge cutoff moves from 20 April to 30 April 2026.²²³³

Reasoning effort runs from low to max, with medium as the default. The values none and minimal are rejected.³³ Tools need the Responses API. Chat Completions takes only requests without them.³³ A multi-agent mode, in beta, lets the model hand work to subagents inside one Responses request.¹¹

OpenAI rates GPT-6.1 Sol Critical in cybersecurity and High in biological and chemical capability.⁴⁴ GPT-6 Sol was High in both.⁶⁶6OpenAI, GPT-6 Astra System Card, HealthBench section and appendix on GPT-6 Sol and GPT-6 Luna, added 22 September 2026. https://deploymentsafety.openai.com/gpt-6-astra The new model runs behind the same safeguards stack as Astra.⁴⁴

OpenAI calls its capabilities comparable to Astra's.⁴⁴ On HealthBench Professional, length-adjusted, the addendum gives it 64.2, against 64.7 for Astra and 60.8 for GPT-6 Sol.⁴⁴ Across all four HealthBench tests its length-adjusted scores stay within 0.5 points of Astra's.⁴⁴

Using it

In the API the model is gpt-6.1-sol.³³

Up to 272K input tokens, Standard pricing is $2 per million input tokens and $10 per million output tokens.⁵⁵ Cached input costs $0.10 and a cache write $2.50.⁵⁵ Above 272K, input doubles to $4 and output rises by half, to $15.⁵⁵ Both rates are a fifth of Astra's.⁵⁵

Batch and Flex cost half of Standard, and Fast mode costs double.³³ US and EU data residency are both supported, but Fast mode is not available with the EU option.³³

OpenAI's model selection guide suggests it for complex projects where cost matters. Its examples are a board presentation built from financial results and a website built from a product brief.⁷⁷7OpenAI, model selection guide, read 29 September 2026. https://developers.openai.com/api/docs/guides/model-selection The same guide says to run it against Astra on one task and weigh quality against cost.⁷⁷