Claude’s New Flagship Pulled After 3 Days, While China’s GLM‑5.2 Opens Instantly

Claude Fable 5 withdrawn three days after launch

On June 9 Anthropic released Claude Fable 5 and Mythos 5. Six days later, on June 13, Anthropic announced a regulatory directive that required an immediate halt to all customer access to both models, affecting every client and even some internal staff. The shutdown was not a regional limitation or a temporary plan outage; it applied globally.

Anthropic explained that regulators were concerned about a potential method to bypass Fable 5’s safety guard. After reviewing the demonstration, Anthropic judged the issue to be a minor vulnerability that also existed in other public models without safety bypasses.

The company argued that recalling a widely deployed model over a single limited bypass would create high uncertainty for future frontier‑model releases.

The abrupt removal left developers who had just integrated Fable 5 into their workflows without any grace period or migration window. Developer Theo remarked that “the model’s capability is impressive, but unstable access makes production risky,” underscoring a new risk for AI‑powered coding tools.

GLM‑5.2 opened to all Coding Plan tiers hours later

At 15:56 on June 13, Zhipu announced that GLM‑5.2 was now available to every Coding Plan tier—Lite, Pro, Max, and Team.

GLM‑5.2 is positioned for coding agents and long‑running tasks, offering:

1 M context : use glm-5.2[1m] in Claude Code with a compression window to enable million‑token context.

Higher inference intensity : the xhigh, max, and ultracode settings in Claude Code map to the max effort of GLM‑5.2, recommended for complex coding jobs.

Multiple agent integration : official docs provide switch methods for Claude Code, OpenClaw, and Cline.

Open plan : the API and chatbot services are slated for release next week, and the model will be offered under an MIT license.

Previously GLM‑5 was limited to Max users due to compute constraints; GLM‑5.2’s full‑tier rollout moves it from a niche preview to large‑scale usage.

For front‑end and Node.js developers, the transition cost is low if they already use Claude‑Code‑style agents—changing an environment variable and model name is often sufficient, leaving the agent shell, project rules, and tool‑call flow intact while avoiding vendor lock‑in.

One model out, another in: shifting competition in AI coding

The sudden removal of a freshly upgraded closed‑source model and the rapid expansion of a compatible, openly licensed alternative illustrate that competition among AI coding models now hinges on openness, integration cost, and service stability rather than raw capability alone.

While GLM‑5.2’s real‑world performance still requires validation, the Fable 5 shutdown reminds developers that even the most capable models need a fallback option, a consideration that may outweigh a single benchmark lead.