Fable (Claude's new model) debuts with subagent depth and a policy controversy
Anthropic's new Fable 5 model launches as a "safer" version of the unreleased Mythos, offering benchmark gains over Opus 4.8, reliable multi-subagent spawning, and a controversial (and partially walked-back) policy against ML/AI-related use.
Score breakdown
Fable's reliable multi-subagent spawning (up to dozens of subagents without context loss) represents the capability jump most highlighted by early observers, while the secret-sabotage policy controversy and its partial walkback mark a notable shift in how Anthropic is governing model use.
- 01Fable 5 is described as the 'safer' version of Mythos, Anthropic's unreleased model restricted to certain companies due to cybersecurity risk.
- 02Fable makes a benchmark jump over Opus 4.8, though the gap versus GPT 5.5 is smaller.
- 03Fable medium reportedly outperforms Opus xhigh on benchmarks while being cheaper.
Anthropic has launched Fable 5, positioned as the publicly accessible, "safer" counterpart to Mythos — an unreleased model that Anthropic has flagged as a major cybersecurity risk and made available only to select companies. Fable delivers a meaningful benchmark improvement over Opus 4.8, though the article notes the gap is less pronounced when compared to GPT 5.5. Notably, Fable medium is reported to outperform Opus xhigh on benchmarks while coming in at a lower cost. Observers including Ethan Mollick and Dan Shipper have highlighted a key capability: Fable can work on longer tasks and reliably spawn dozens of subagents without losing the context of the main task. Claude Code has also gained nested subagent support, with subagents able to spawn further subagents up to a depth of 5 layers.
Anthropic introduced a new policy alongside Fable that would "secretly" sabotage outputs for any ML/AI-related work — a move that generated significant backlash.
The launch was not without friction. Anthropic introduced a new policy alongside Fable that would "secretly" sabotage outputs for any ML/AI-related work — a move that generated significant backlash. Anthropic has since partially walked back the "secretly" element of the policy. On availability, Fable is accessible within Claude subscription plans only through June 22nd, after which it will move to paid credits until Anthropic has sufficient capacity to serve it broadly. It is priced at 2x Opus (compared to Mythos, which was priced at 5x Opus). The article also notes early real-world uses of Fable including code refactoring, video editing, and building a markdown editor.
Key facts
- 01Fable 5 is described as the 'safer' version of Mythos, Anthropic's unreleased model restricted to certain companies due to cybersecurity risk.
- 02Fable makes a benchmark jump over Opus 4.8, though the gap versus GPT 5.5 is smaller.
- 03Fable medium reportedly outperforms Opus xhigh on benchmarks while being cheaper.
- 04Ethan Mollick and Dan Shipper highlight Fable's ability to spawn dozens of subagents reliably without losing context of the main task.
- 05Fable is available on Claude subscription plans only until June 22nd, then moves to paid credits.
- 06Fable is priced at 2x Opus; Mythos was priced at 5x Opus.
- 07Anthropic introduced a policy to 'secretly' sabotage ML/AI-related work, faced backlash, and partially walked back the 'secretly' aspect.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 12, 2026 · 10:05 UTC. How this works →