Claude Fable 5 and Mythos 5 Released: Anthropic Unlocks the Ultimate AI Tier
Anthropic has officially launched Claude Fable 5 and Claude Mythos 5. Learn about the mind-bending agentic capabilities, the Opus 4.8 fallback safety system, and why the model is so powerful it requires a dual release.
TL;DR
Anthropic just launched Claude Fable 5 and Claude Mythos 5. These models sit in an entirely new class above the Opus tier and represent the most capable AI systems ever deployed to the public.
Fable 5 is built for general users and features aggressive safety constraints. Mythos 5 is the exact identical model with safeguards removed, restricted strictly to government and defense partners under Project Glasswing.
If you ask Fable 5 a question about cybersecurity or biology, its safety classifiers block the request. Instead of giving you a hard refusal, it automatically falls back to Claude Opus 4.8 to answer your prompt.
Fable 5 beat Pokémon FireRed using only raw visual screenshots. It autonomously built an automated factory in Factorio and compressed months of complex engineering at Stripe into just a single day.
Because the model requires massive compute, Fable 5 is only included in standard subscription plans until June 22, 2026. After that date, it will require pay-as-you-go usage credits until Anthropic scales its server capacity.
Claude Fable 5 and Mythos 5: Anthropic Unlocks the Ultimate AI Tier
Anthropic has finally opened the vault. After months of intense speculation surrounding the leaked "Capybara" project and the highly restricted Project Glasswing, the AI world finally gets to touch the next generation of intelligence.
On June 9, 2026, Anthropic announced the launch of Claude Fable 5 and Claude Mythos 5.
This release fundamentally changes the global AI leaderboard. By introducing an entirely new "Mythos" class that sits high above the already dominant Opus tier, Anthropic has unleashed a system with capabilities that border on science fiction. However, they are also introducing a radically new approach to AI safety to prevent this monumental power from being misused.
Here is a complete breakdown of the mind-bending agentic capabilities, the fascinating Opus 4.8 fallback system, and the bizarre subscription rules you need to know before you start prompting.
The Dual Release Strategy
Anthropic faced a massive dilemma earlier this year. Their newest model was so incredibly intelligent that red-teaming revealed terrifying offensive cybersecurity and bioweapon capabilities. Their solution was to split the model into two distinct entities.
Claude Fable 5 is the public-facing model. It is designed for general use and is available right now. It possesses the full intellectual horsepower of the new architecture but is wrapped in highly aggressive safety classifiers.
Claude Mythos 5 is the exact same underlying neural network. The key difference is that its safeguards are lifted. Because of the extreme dual-use risk, Mythos 5 remains strictly locked behind Project Glasswing. It is only accessible to a trusted access program comprising US government agencies, elite cyber defenders, and a small cohort of molecular biology researchers.
The Brilliant Opus 4.8 Fallback System
One of the most fascinating aspects of Fable 5 is how it handles refusals. Historically, when an AI model flagged a prompt as dangerous, it simply apologized and refused to answer. Anthropic realized this was deeply frustrating for legitimate users caught by false positives in the safety filters.
With Fable 5, Anthropic introduced the intelligent fallback system.
If the safety classifiers detect a request related to cybersecurity, advanced biology, chemistry, or model distillation, Fable 5 will refuse to process it. But instead of generating a useless error message, the system automatically routes your exact prompt to Claude Opus 4.8.
Because Opus 4.8 is currently the top-ranked model on the global Artificial Analysis leaderboard, users still receive a brilliant, world-class answer without being blocked. Anthropic noted that these safety fallbacks currently trigger in less than 5 percent of sessions, meaning the vast majority of users will experience the pure, unadulterated power of Fable 5.
Mind-Bending Capabilities and Benchmarks
The benchmark scores for Fable 5 are staggering. It sets a new state-of-the-art ceiling across software engineering, vision, and knowledge work. But the real magic lies in its autonomous agentic behavior.
1. Gaming Through Pure Vision
Here is how Claude Mythos 5 and Fable 5 stack up against the competition across key frontier benchmarks:
| Benchmark | Claude Mythos 5 / Fable 5 | Claude Opus 4.8 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|
| Agentic Coding (SWE-Bench Pro) | 80.3% | 69.2% | 58.6% | 54.2% |
| Agentic Coding (FrontierCode Diamond) | 29.3% | 13.4% | 5.7% | — |
| Knowledge Work (GDPval-AA) | 1932 | 1890 | 1769 | 1314 |
| Computer Use (OSWorld-Verified) | 85.0% | 83.4% | 78.7% | 76.2% |
| Spatial Reasoning (Blueprint-Bench 2) | 38.6% | 14.5% | 36.2% | 26.5% |
| Multidisciplinary (Humanity's Last Exam - with tools) | 64.5%* | 57.9% | 52.2% | 51.4% |
| Cybersecurity (ExploitBench Cap%) | 78.0%* | 40.0% | 34.0% | — |
| Biology (BioMysteryBench - human solved) | 83.9%* | 80.4% | — | — |
*Note: Starred benchmarks show the raw score of Mythos 5. Because Fable 5 utilizes strict safety blockers for cybersecurity and biology questions, it will automatically fall back to Opus 4.8 on these specific prompts, resulting in Opus-level effective scores during those specific public usage scenarios.
Previous AI models required complex textual harnesses and API hooks to understand and play video games. Fable 5 changes everything. Anthropic showcased the model playing Pokémon FireRed from start to finish using absolutely nothing but raw visual screenshots.
It did not stop there. Fable 5 autonomously built a highly complex automated factory in the engineering game Factorio. It even used persistent file-based memory to navigate Slay the Spire, reaching the final act three times more often than Opus 4.8.
2. Enterprise Code Migration
During early beta testing, the financial technology giant Stripe granted Fable 5 access to a massive 50-million-line Ruby codebase. The model executed a complete, codebase-wide migration in a single day. A human engineering team would have taken over two months to complete the exact same task by hand.
3. Scientific Breakthroughs
Using the unrestricted Mythos 5 variant, Anthropic scientists generated entirely novel hypotheses in molecular biology. In one highly publicized case, a novel mechanism proposed by Mythos 5 for an E. coli protein was corroborated weeks later by an independent human laboratory working on the exact same problem.
It also assembled single-cell genomics data spanning 138 animal species, training a custom machine learning model that outperformed a recent model published in the journal Science while being 100 times smaller.
Pricing and The Capacity Crunch
Operating a model of this magnitude requires an unbelievable amount of data center compute. Through the API, Fable 5 is positioned as a hyper-premium product. It is priced at $10 per million input tokens and $50 per million output tokens. This places it firmly at the top of the market in terms of cost.
For consumer and enterprise subscribers using the Claude web console, there is a major catch regarding availability.
From today until June 22, 2026, Fable 5 is included for free on Pro, Max, Team, and Enterprise subscription plans. However, on June 23, Anthropic will physically remove Fable 5 from the standard subscription pool. Using it after that date will require purchasing specific pay-as-you-go usage credits. Anthropic confirmed this unusual step is necessary until they can deploy enough server capacity to handle the overwhelming global demand.
Additionally, Anthropic instituted a strict 30-day data retention policy for all traffic on Mythos-class models to monitor for sophisticated jailbreaks and advanced misuse.
The Verdict
The launch of Claude Fable 5 proves that the intelligence ceiling of artificial intelligence has not even come close to being reached.
By creating a model so powerful that it requires a dual release strategy and an entirely new fallback safety net, Anthropic has redefined what a frontier AI system looks like. While the strict pricing and temporary subscription access might frustrate some early adopters, the pure capability of Fable 5 makes it the undeniable apex predator of the AI ecosystem in June 2026.
Compare Fable 5 against GPT-5.5 and DeepSeek V4 in our tools directory →
Frequently Asked Questions
What is the difference between Claude Fable 5 and Claude Mythos 5?
They are the exact same underlying neural network. Fable 5 is the public version equipped with strict safety classifiers that block sensitive queries. Mythos 5 has these safeguards completely removed and is exclusively available to defense partners and select molecular biology researchers.
How does the Opus 4.8 fallback work in Fable 5?
If you ask Fable 5 to perform a task related to offensive cybersecurity, chemical engineering, or model distillation, the safety classifier will intercept the prompt. Instead of giving you an annoying error message, Anthropic automatically routes your prompt to Claude Opus 4.8, ensuring you still get a world-class response.
How much does Claude Fable 5 cost?
Through the API, Fable 5 costs $10 per million input tokens and $50 per million output tokens. For web console users, it is included in Pro and Enterprise plans until June 22, 2026. After that date, it will require usage credits due to severe compute constraints.
Can Claude Fable 5 play video games?
Yes. Anthropic demonstrated Fable 5 playing Pokémon FireRed from start to finish using only raw visual screenshots with no extra code scaffolding. It also autonomously played the complex engineering game Factorio and used persistent memory files to master Slay the Spire.
