Week of Sep 7
← Week of Aug 31Week of Sep 14 →
What changed (from public sources)
OpenAI opens a Codex agent API, and DeepSeek releases V4.1-Flash
On September 10, DeepSeek released V4.1-Flash, the smallest model in its new architecture family, and said tests by multiple parties put it ahead of its own flagship V4-Pro [3]. The same day OpenAI opened the Agents API, a managed Codex harness, to all developers in public beta [1][2].
OpenAI turns the Codex harness into a public-beta Agents API
OpenAI released the Agents API in public beta on September 10: agents run on a managed
Codex harness while OpenAI handles session orchestration, context compaction and recovery [1]. Agents can run in OpenAI-hosted sandboxes or in sandboxes from partners including Cloudflare, Daytona, E2B, Modal and Vercel, and there is no separate fee beyond tokens and tools [2].
552B
parameters in the V4.1-Flash MoE [3]
DeepSeek V4.1-Flash arrives and takes over V4-Pro API traffic
DeepSeek introduced V4.1-Flash on September 10, a 552B-parameter MoE with native visual understanding and 8B active parameters for input [3]. DeepSeek says it is phasing out V4-Pro: from 04:00 UTC on September 14, deepseek-v4-pro requests route to V4.1-Flash at V4.1-Flash rates, and it lists opencode among partners that fully support the new model [3]. The claim that it beats V4-Pro is DeepSeek's own summary of outside tests.
News this week
ChatGPT Changed: OpenAI plans to retire custom GPTs in ChatGPT with a migration path to plugins
Cursor Cursor launches Projects for long-running multi-agent work
DeepSeek V4 Pro Changed: DeepSeek keeps the V4 Pro API running after September 14 with unchanged billing
Krea Krea Agent is now available on all Krea plans, including free
Notion AI Changed: Notion lets workspace owners choose which AI models agents can use
Suno Suno releases its v6 model family, built with Warner Music, BMG and Believe
Devin Changed: Devin Desktop drops its Cascade agent; Devin Local is now the only one