AI
After US bans Claude Fable, open-source models become real fallback option for enterprises
When the US government suspended Anthropic's flagship model, competing open-weight coding systems rushed in—showing how quickly enterprises can swap models when one hosted service disappears.
The New Stack
On June 12, the US government ordered Anthropic to suspend Claude Fable 5 and Mythos 5 for all foreign nationals. The company pulled both models for all customers worldwide, citing a suspected jailbreak vulnerability. By week's end, three major open-weight coding models had stepped in as viable alternatives—some already in development before the ban, others timed to capitalize on the sudden demand.
The sudden vacuum#
Fable 5 had launched just nine days earlier on June 9, claiming benchmarks that positioned it as Anthropic's strongest model yet. Any enterprise automation built on it lost access in an afternoon.
Between June 9 and June 13, four open-weight models became real options:
- North Mini Code (Cohere, June 9): A 30-billion-parameter sparse model that runs on a single H100. Apache 2.0 licensed.
- Kimi K2.7-Code (Moonshot, June 12): 1 trillion parameters with downloadable weights; tuned for multi-file coding tasks.
- GLM 5.2 (Zhipu, June 13): Gated behind a paid plan with a one-million-token context window. Open weights promised to follow.
- MiMo-V2.5-Pro (Xiaomi, April): Earlier entrant, full weights available.
GLM 5.2 drew particular attention when a community leaderboard (BridgeBench) placed it at the top—though the claim remains unverified pending independent testing and weight release.
Why enterprises can actually switch#
The speed of fallback mattered because the tooling already supports it. Claude Code, Cursor, and other agentic editors converged on a shared protocol layer; swapping model endpoints no longer requires rebuilding workflows. A team running GLM 5.2 through Claude Code simply changes where it points—the orchestration layer stays intact.
None of these models were built in response to the ban. North Mini Code and Kimi K2.7-Code shipped before or on the same day as the order. But the ban turned what would have been routine releases into a critical second-source hedge.
What comes next#
Zhipu's move to release GLM 5.2 weights this week will test whether the open alternative is genuinely competitive or merely cheaper. Independent benchmarking under real conditions will settle whether enterprises see this as a sustained shift or an interim patch.
