Summary
OpenAI’s DevDay brought a long list of announcements, and Sam Witteveen goes through the ones that matter most to developers, with his own take on each.
The headline item is Dots, an always-on agent available inside ChatGPT and Codex, with its own cloud computer and browser, access to the plugins and connectors already linked to an account, and Slack integration. Witteveen compares it with OpenClaw, Hermes and Gemini Spark, and explains the usage model: talking to a dot is included in a plan, while the work it performs counts toward token usage. He also speculates about how it could tie into OpenAI’s upcoming hardware.
On the model side, he covers GPT 6.1 Soul, positioned as close to the flagship Astra at roughly a fifth of the cost: $2 per million input tokens and $10 per million output tokens, versus $10 and $50 for Astra. A 95% discount on cached input makes the cache price the effective price for agent workloads, and he shares that about 96% of 250 million tokens in one of his own Opus 5.5 projects were cached.
The video also covers Codex updates, including Linux support, multiple folders per project, Codex in the cloud and on phones, and bidirectional voice powered by GPT Live.
šŗ Source: Sam Witteveen Ā· Published September 30, 2026
š·ļø Format: News Analysis
