June 9, 2026Anthropic Dropped a bombshell: Released Claude Fable 5——This is Anthropic Mythos-class model is available to the public for the first time. Claude Opus 4.8.
core competencies
Fable 5 is based on the new Mythos architecture, which Anthropic calls "the most powerful model ever trained" in the system card.
- Magenta Code Benchmark:Fable 5 score 80%, about 6 times that of GPT-5.5, showing an overwhelming advantage in programming agent tasks
- Comprehensive benchmark: More than an improvement over Opus 4.8 10%, reaching or approaching the state-of-the-art level in almost all testing areas
- long range mission: Specially optimized for "complex tasks where early models are easy to give up midway", supporting ultra-long context windows (1M tokens)
- Agentic capabilities: Amazing performance in complex agent tasks such as Pokémon FireRed clearance and solar system simulation.
In actual experience, Fable 5’s performance in programming scenarios is indeed impressive.
Pricing and availability
The pricing of Fable 5 is positioned in the high-end market:$10/million input token, $50/million output token- About 60% more expensive than the Opus 4.8. Free use window until June 22, which allows users who want to try it out to experience it at zero cost.
It is worth noting that Fable 5 introduces new 30-day data retention requirement, which is different from Anthropic’s usual privacy-friendly policy, and enterprise users need to pay attention to this change.Claude Code and web use.
User experience/limitations
- advantage: Qualitative improvement in programming ability, excellent long-term task stability, and excellent creative writing performance
- shortcoming: Expensive ($50/million output tokens), the security guardrails are too conservative (basic biological questions are also blocked), and the 30-day data retention policy may deter privacy-sensitive users.
- Applicable scenarios: Complex programming projects, research-level knowledge work, long-form writing requiring in-depth reasoning
- Not applicable scenarios: High-frequency low-cost calls, biology/chemistry related research, and zero tolerance for data privacy scenarios
Overall Score
| 维度 | Score | evaluate |
|---|---|---|
| functional completeness | 9.5 / 10 | Comprehensive leadership in programming reasoning, complete multi-modal support |
| 易用性 | 8.0 / 10 | Safety guardrails are too aggressive in some scenarios |
| Cost-effectiveness | 7.0 / 10 | The output cost is high, but Pro/Max is free before June 22 |
| 中文支持 | 8.5 / 10 | Can understand Chinese fluently, but not as authentic as domestic models |
| 输出质量 | 9.5 / 10 | Programming and reasoning quality reaches new heights |
Overall rating: 8.7/10
Claude Fable 5 is Anthropic’s answer to the market on the eve of its IPO: it proves that Anthropic can not only be “safe” but also “strongest”. AI Dash ——Discover the best AI tools.