OpenAI raises default context for 'GPT-6.1 Sol' in Codex and ChatGPT Work from 272K to 500K tokens (report)

Verified (fact)

According to AI Times, OpenAI product lead Thibault Sottiaux announced the change on the 12th (local time). The default context length of GPT-6.1 Sol, used in Codex and ChatGPT Work, increased from 272K tokens to 500K tokens, about 80% more. It is the day-7 update of OpenAI's ongoing '28 consecutive days of Codex performance improvements' event.

Needs fact-check (claim)

According to OpenAI, long-running tasks will require less context compaction, and details of complex tasks can be retained for longer.

Why it matters (interpretation)

For long-running work such as coding agents, the context limit is a key constraint on performance. Because the default changed without any extra configuration, existing users are affected immediately.

Sources

• Press 오픈AI, 'GPT-6.1 솔' 컨텍스트 길이 50만 토큰으로 대폭 상향 AI Times (Korea)

More about OpenAI

• 2026-10-13 P2 Model Unconfirmed Generative media OpenAI
  OpenAI in talks to license its shut-down video model Sora to a startup co-founded by Katzenberg, in exchange for equity (report)

  AI Times reported this, citing The Information. OpenAI, which shut down development and service of Sora due to operating costs and...

• 2026-10-12 P1 Policy Unconfirmed Safety & alignment US administrationOpenAIAnthropic
  Trump administration reverses course, lets US AI firms give UK AI Security Institute pre-release model access (report)

  AI Times, citing a Telegraph report from Oct. 11 (local time), said the Trump administration has reversed its earlier position that told...

• 2026-10-12 P2 Research Paper Safety & alignmentLanguage models Carnegie Mellon UniversityUC BerkeleyMITUniversity of OxfordUCLAOpenAI
  Study: just 10 minutes of AI use makes people give up more easily once help is removed; 3 RCTs, 1,222 participants (report)

  Researchers led by first author Grace Liu of Carnegie Mellon University, with UC Berkeley's Center for Human-Compatible AI (CHAI), MIT,...

• 2026-10-12 P2 Model Unconfirmed Science & bioLanguage models OpenAI
  OpenAI life sciences model 'gpt-rosalind-discovery' appears on API pricing page before any announcement (report)

  ZDNet Korea, citing TestingCatalog, reported that 'gpt-rosalind-discovery' was added to the life sciences model section of OpenAI's API...

• 2026-10-12 P0 Model Generally available Language modelsAgents MetaAnthropicOpenAI
  Meta unveils first frontier models Muse Wonder and Muse Wonder Full, claims lead over Opus 5.5 on agent and coding benchmarks (reports)

  According to TechCrunch and The Verge, Meta has unveiled Muse Wonder and Muse Wonder Full, new models from Meta Superintelligence Labs....