GPT-5.6 and ChatGPT Work: OpenAI drops a workhorse, Anthropic answers with a reset

OpenAI ships GPT-5.6, folds Codex into ChatGPT Work for apps and files, and cuts the top-tier price in half. Anthropic responds the same day by wiping Claude's usage limits. The frontier war is now billed by the meter.

In plain terms

OpenAI just released GPT-5.6 with a repackaged desktop app called ChatGPT Work, which merges the Codex coding agent with document and app access. Anthropic answered the same day by resetting every Claude user's monthly usage limit. Two frontier vendors, one message: the ceiling is compute, and both know it.

The move

GPT-5.6 is positioned as OpenAI's new default. According to Inc42, the top-tier model claims "half the cost with parts exceeding Fable 5 in benchmarks" - a rare admission that pricing, not raw capability, is now the lever. ChatGPT Work bundles Codex (the coding agent) with file and app connectors: it is a direct answer to Claude Cowork, launched by Anthropic weeks earlier and already integrated into Microsoft's M365 Copilot stack.

The Codex/ChatGPT Work usage-limit reset - followed within hours by Claude's own limit wipe - was flagged by ITmedia. An OpenAI executive publicly needled Anthropic: "they're scared." That is the tone of a market that has stopped competing on model IQ and started competing on how much of it you can burn per day.

Under the hood

The pattern under the noise:

• Model layer: no gap left to open on generic benchmarks - GPT-5.6 leads on some tasks, trails on others, and prices aggressively.
• Agent layer: both vendors now default-bundle a coding agent (Codex / Claude Code) with a document/app surface.
• Distribution layer: Microsoft ships GPT-5.6 in M365 Copilot; Anthropic ships Claude natively in Bedrock and Notion. Neither vendor still wants to be only a model API.

Read this as product convergence, not model convergence. The differentiators are latency, tool-use reliability, and how many tokens users burn before hitting a wall.

So what

For builders: expect the "per-message limit" to keep swinging as both sides trade PR punches. Don't rearchitect around a limit that resets weekly. For decision-makers: the price of a Claude/GPT seat is now the compute you can extract before the meter cuts you off - not the model name on the sticker. And for anyone paying enterprise contracts: renegotiate. The floor is dropping fast.