The official version of 'Claude Mythos' has finally been released, along with 'Claude Fable,' a version with no user restrictions, making it available to everyone.
AI company Anthropic released ' Claude Mythos 5 ' and ' Claude Fable 5 ' on June 9, 2026, local time. Both models have equivalent capabilities; Claude Mythos 5 is a security-removed version released exclusively to select organizations, while Claude Fable 5 is a retail version with security measures in place and will be released to the public.
Claude Mythos \ Anthropic
Claude Fable \ Anthropic
https://www.anthropic.com/claude/fable
Claude Fable 5 and Claude Mythos 5 \ Anthropic
https://www.anthropic.com/news/claude-fable-5-mythos-5
On April 7, 2026, Anthropic announced ' Claude Mythos Preview ,' a high-performance AI model capable of sophisticated cyberattacks, and has been providing it to select organizations with the aim of strengthening their cyber defense capabilities. Claude Mythos 5 and Claude Fable 5 are positioned as updated versions of Claude Mythos Preview and are touted as having the highest performance among the AI models Anthropic has released to date.
Anthropic offers a lineup of consumer models including the lightweight and fast 'Haiku,' the balanced 'Sonnet,' and the high-performance 'Opus,' with the 'Fable' now added as the top-of-the-line model. Incidentally, Haiku means 'haiku,' Sonnet means '14-line poem,' Opus means 'work/masterpiece,' Mythos means 'myth/legend,' and Fable means 'myth/legend/fable.'
Claude Mythos 5 and Claude Fable 5 are identical models except for the presence or absence of security restrictions. Below is a table comparing the benchmark performance of 'Claude Mythos 5 and Claude Fable 5,' 'Claude Mythos Preview,' 'Claude Opus 4.8,' 'GPT-5.5,' and 'Gemini 3.1 Pro.' Claude Mythos 5 and Claude Fable 5 record scores that significantly outperform high-performance models from OpenAI and Google.
The graph below compares the performance of 'Claude Fable 5,' 'Claude Opus 4.8,' and 'GPT-5.5' as coding agents. Claude Fable 5 significantly outperforms Claude Opus 4.8, and there is a considerable difference between it and GPT-5.5.
Claude Fable 5 excels particularly in long-term autonomous tasks. Stripe, which participated in the initial testing, reported that 'Claude Fable 5 was able to complete a task involving 50 million lines of Ruby code in one day, a task that would have taken the entire team more than two months to complete.' AI researcher
The graph below shows the performance increase rate relative to cost. In Claude Opus 4.8 and GPT-5.5, performance plateaued even as cost increased, but in Claude Fable 5, performance improved as the cost increased.
A video has also been released showing how to complete 'Pokémon FireRed' using Claude Fable 5.
Claude Fable 5 beats Pokémon FireRed only using vision - YouTube
Claude Fable 5 includes security measures to reject instructions related to cyberattacks, and for some topics, it employs a mechanism where Claude Opus 4.8 will respond instead. Restrictions have been tightened to ensure safe and rapid model releases, and it is estimated that approximately 5% of all sessions will switch to responses from Claude Opus 4.8.
The graph below compares the cyberattack capabilities of 'Claude Opus 4.8,' 'Claude Mythos Preview,' 'Claude Mythos 5,' and 'Claude Fable 5.' In this comparison experiment, Claude Fable 5 was used in a mode that 'interrupts the response rather than leaving it to Claude Opus 4.8.' As a result, while Claude Mythos 5, with the same performance, was able to execute a large number of attacks, the attack success rate for Claude Fable 5 was 0%.
The API fees for Claude Mythos 5 and Claude Fable 5 are $10 (approximately 1600 yen) per 1 million input tokens and $50 (approximately 8000 yen) per 1 million output tokens. Claude Mythos 5 is only available to select organizations.
Related Posts:
in AI, Posted by log1o_hf