Anthropic just launched Claude Fable 5, its first Mythos-class AI model – but it has new safeguards to prevent misuse and will ‘fall back’ to Opus 4.8 for ‘high risk’ queries
The launch of Claude Fable 5 marks the first public release of a Mythos-class AI model
Anthropic has unveiled a new Mythos-class AI model designed specifically for public use.
The new model, dubbed Fable 5, boasts significant performance improvements on previous models, with the company noting that it “exceeds every model” previously made available to the public.
Notably, Fable 5 has been tweaked to include new safeguards that “block responses in specific high-risk areas” to prevent misuse – a key concern with the launch of Claude Mythos.
Claude Mythos was rolled-out as part of a gated release in April this year, with industry partners such as Microsoft, Amazon, and Cisco selected to test the model’s capabilities behind closed doors.
With the launch of Fable 5, Anthropic revealed that the model will “fall back” to Claude Opus 4.8 in the event that potential high-risk uses are detected.
Alongside the launch of Fable 5, Anthropic also revealed it will release Claude Mythos 5, again in a restricted capacity.
“Beginning today, all users who currently have access to Claude Mythos Preview (for example, our cybersecurity partners in Project Glasswing) will be able to upgrade to Claude Mythos 5,” the company said in an announcement.
The company added that it plans to expand access in due course through a “more systematic trusted-access program”.
Performance boosts with Claude Fable 5
According to Anthropic, Fable 5 shows “exceptional performance” in software engineering-related tasks, as well as knowledge work and scientific research.
Testing of the model by analytics company Hex revealed the model was the first to record a score of over 90% on its core analytics benchmark, which consists of running the model through “complex, long-running analytical tasks”.
Indeed, Hex noted that Fable recorded a 10-point improvement over Opus.
Similar testing by agentic workspace platform, Genspark, also showed significant performance boosts compared to other models, particularly in areas such as coding.
“Fable came out #1 on our evals, winning head-to-head against every model we tested,” the company said. “It was significantly stronger on the hardest tasks in the set — UI design and game coding."
Robust guardrails for Fable 5
With Fable 5’s new safeguards, the model will be restricted from performing tasks in a range of areas, including cybersecurity and biology, and will revert back to Opus 4.8 to provide answers to queries on these topics.
“Releasing a model this capable comes with risks,” Anthropic said in a statement. “Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage.”
Anthropic said it “extensively red-teamed” classifiers to test the model’s defenses against potential jailbreaks as part of a bug bounty.
“Internally, we ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing,” the company said. “We then worked with external red-teaming orgs which also failed to find universal jailbreaks.”
Fable 5 pricing and access
According to Anthropic, pricing for both models will stand at $10 per million input tokens, with $50 per million output tokens.
Token consumption costs have become a hot topic in the tech industry over the last two months with the rise of ‘tokenmaxxing’.
Uber, for example, revealed it blew through its entire annual AI budget in just four months amidst skyrocketing use by staff.
Developers will be able to use Fable 5 through the Claude API while Mythos 5 will only be accessible to Project Glasswing participants.
FOLLOW US ON SOCIAL MEDIA
Follow ITPro on Google News and add us as a preferred source to keep tabs on all our latest news, analysis, views, and reviews.
You can also follow ITPro on LinkedIn, X, Facebook, and BlueSky.
-
The next IoT challenge: making connected products work at global scaleAs connected products move from pilots to operational infrastructure, organizations focused on scaling IoT need to think beyond the device and design for reliability, security and scale from day one
-
Meta is launching a trades academy to bolster AI infrastructure build-outsNews The tech giant will stump up to help train workers who will build data centers – and guarantees them jobs
-
Nine seconds was all it took for an AI agent to wipe a startup’s database —experts warn it’s a glimpse into the future challenges of identity securityNews The recent PocketOS incident shows the growing identity security risks associated with AI agents, according to cyber experts
-
‘We experimented with efforts to differentially reduce these capabilities’: Anthropic toned down Opus 4.7’s cyber uses in wake of Claude Mythos releaseNews Testing of new cyber-related safeguards for Anthropic’s Opus 4.7 model could shape the future public release of Claude Mythos
-
Anthropic is worried hackers could abuse its Claude Mythos AI model – so it's asking big tech partners to test it behind closed doorsNews Anthropic’s Project Glasswing will give a host of leading tech companies access to its new Claude Mythos model for testing
-
Microsoft is rolling out Copilot Cowork to more customersNews Use of Copilot Cowork has been limited to select customers so far
-
Microsoft has a new AI poster child in Anthropic – and it’s about timeOpinion Microsoft is cosying up to Anthropic at a crucial time in the race to deliver on AI promises
-
Anthropic's Claude Cowork tool is coming to Microsoft CopilotNews The new Copilot Cowork tool will be made available through a new Microsoft 365 tier at the end of March
-
Will AI hiring entrench gender bias?ITPro Podcast This International Women's Day, it's more important than ever to consider the inherent biases of training data
-
February rundown: SaaS-pocalypse now?ITPro Podcast Geopolitical uncertainty is intensifying public and private sector focus on true sovereign workloads
