OpenAI released GPT-6.1 Sol at DevDay on September 29 at one-fifth of GPT-6 Astra's price with a 95% cached-read discount, sitting between GPT-6 Astra and GPT-6 Luna in capability.
A system card addendum shows HealthBench Professional at 64.2 and an ExploitBench arbitrary code-execution rate of 21.5%, above prior Sol models but below GPT-6 Astra's 31.5%. OpenAI also launched an Ultrafast tier capped at roughly 300 tokens per second for 6x standard rates; tool calling is limited to the Responses API and ChatGPT Chat access was not yet available at launch.