Developers run Claude Code on non-Anthropic models — Anthropic loses the token bill
What's the deal? The Information reports developers are pairing AnthropicDealroom has a profile for this one. Try Dealroom →’s Claude Code coding harness with cheaper non-Anthropic models — including OpenAI’s GPT-5.6 Sol — via proxies and routers to cut costs. When Claude Code runs on someone else’s model, Anthropic loses the inference revenue.
Caveats / inference: Manufacturer-AI founder Adam Kirsh’s ~97% quality / ~60% lower cost claim (plan on a frontier model, execute on cheaper open models) is a founder estimate, not a measured Anthropic revenue leak. TI gives no share of Claude Code users on third-party models and no official ToS enforcement policy beyond the spokesperson line. Proxy crackdowns may be uneven.
The signal: Coding agents are splitting into a harness layer and a token layer. If the UX can ride another lab’s model, labs that priced on locked-in token spend face a quiet GTM leak — even when they “win” the developer interface.
Source: The Information — Developers Find Ways to Use Claude Code Without Anthropic Models (Alix Coutures, 15 Sep 2026).
