OpenAI ChatGPT Work 1.0 ships with GPT-5.6; Anthropic J-lens 0.1 maps J-space in Claude 4

ChatGPT Work 1.0 combines conversation, code execution and model routing under one authenticated session. The release coincided with GPT-5.6 parameter counts and benchmark scores listed in OpenAI’s internal changelog. Concurrently Anthropic’s J-lens paper (arXiv:2607.03116) describes a Jacobian-based probe that isolates token subspaces containing response candidates not emitted in the final output. Both moves occurred within 48 hours, tightening control over model access and internal state visibility.

Enterprise telemetry from the ChatGPT Work rollout shows 180 000 new workspace activations in the first 24 hours, per OpenAI’s status dashboard. J-lens 0.1 recovered 47 % of withheld tokens on a 10 k prompt test set drawn from Claude 4’s production logs. These figures indicate measurable expansion of platform surface area rather than isolated model improvements. The pattern matches prior consolidation phases observed after GPT-4 and Claude 3 releases.

Operational effect is reduced switching cost between inference, tool use and state inspection. Developers now route tasks inside ChatGPT Work without separate API keys or external interpretability stacks. Anthropic’s J-space disclosure supplies the first public coordinate system for hidden activations, enabling downstream tooling that locks users deeper into each vendor’s stack. The dual release therefore accelerates vertical integration of generation and observability layers.

Next quarter OpenAI plans to embed the automated researcher agent inside the same workspace; Anthropic has scheduled J-lens 0.2 with gradient routing hooks. Both timelines point to lock-in thresholds at 10 M daily active enterprise seats by December 2026.