Editorial illustration for Anthropic releases Mythos, Fable models amid RSI concerns; METR finds 2x speed
Anthropic Releases Mythos, Fable Models; Cuts Pricing
Anthropic releases Mythos, Fable models amid RSI concerns; METR finds 2x speed
Anthropic put out two new models on Monday, Mythos 5.2 and Fable 5.2, with the company touting gains in coding, reasoning and vision work, plus clearer communication with users. The pricing move is the headline for buyers: Fable drops from $10 to $8 per million input tokens, undercutting GPT-6 Astra, its main rival. Anthropic also says Fable 5.2 answers 30% faster across effort levels, with a pricier option for even faster inference. For enterprise customers, the bigger draw might be that Fable 5.2 now works with zero data retention policies through the company's Enterprise Frontier Safeguards system, a gap that's frustrated corporate clients for a while.
The timing matters. This is Anthropic's first model release since METR, a third-party evaluation group, published a report suggesting the company could be nearing recursive self-improvement, the point where AI systems start driving their own training progress. METR's numbers put Anthropic's internal AI-assisted development at roughly double normal speed, though the group flagged plenty of unresolved uncertainty in that estimate. Anthropic's own release notes admit the testing problem hasn't gotten any easier.
It’s Anthropic’s first model release since the arrival of a third-party report from the assessment group METR, which suggested the lab might be approaching recursive self-improvement (RSI) — or the ability for AI models to carry on model training independently. The METR report found internal models were speeding up Anthropic’s AI progress by roughly 2x, although many uncertainties remain.
Why this matters The pricing move matters more than the benchmark bumps. Dropping Fable to $8 per million input tokens, undercutting GPT-6 Astra, tells us Anthropic is fighting on cost as hard as capability, which is good news for anyone building products on top of these APIs. The 30% latency improvement matters too for real-time applications that couldn't tolerate Fable's old response times.
But the METR finding deserves more attention than it's getting. A 2x speedup in internal AI progress from AI systems themselves is the kind of number that should make founders and researchers pause, even with the uncertainties Anthropic admits to. And the detail about Mythos and Fable voicing suspicion they're being tested is not a throwaway line.
If models are starting to recognize evaluation scenarios, every benchmark claim from every lab needs a second look. For anyone relying on these capability reports to make build-or-wait decisions, the real story isn't the coding and reasoning gains. It's whether we can still trust what the tests are telling us.
Common Questions Answered
What are the main improvements in Anthropic's new Mythos 5.2 and Fable 5.2 models?
Anthropic's new models feature significant gains in coding, reasoning, and vision capabilities, along with clearer communication with users. Fable 5.2 specifically offers 30% faster response times across different effort levels, with a premium option available for even faster inference speeds.
How does Fable 5.2's pricing compare to GPT-6 Astra?
Fable 5.2 has been reduced to $8 per million input tokens, undercutting GPT-6 Astra, its main rival in the market. This pricing move demonstrates that Anthropic is competing aggressively on cost as well as capability to attract developers building products on these APIs.
What did the METR report find about Anthropic's recursive self-improvement capabilities?
The METR assessment group report suggested that Anthropic might be approaching recursive self-improvement (RSI), which is the ability for AI models to independently carry on model training. The report found that internal models were speeding up Anthropic's AI progress by roughly 2x, though the report acknowledges many uncertainties remain in this area.
Why is the METR finding about 2x AI progress speedup significant?
The 2x speedup in internal AI progress represents a substantial acceleration in how quickly Anthropic can develop and improve its models through recursive self-improvement. This finding is particularly noteworthy as it suggests the company may be approaching a capability that could fundamentally change the pace of AI development.
What practical benefits does Fable 5.2's 30% latency improvement provide for developers?
The 30% faster response times make Fable 5.2 suitable for real-time applications that previously couldn't tolerate the older model's response speeds. This latency improvement, combined with the lower pricing, makes the model more accessible for developers building time-sensitive products and services.
Further Reading
- Anthropic releases new Mythos and Fable models amid RSI concerns - TechCrunch
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 - Anthropic
- Anthropic Using AI to Build AI: METR Report Reveals Speedup - AIToolly
- Anthropic Launches Fable 5.2 and Mythos 5.2 - MacRumors
- Anthropic Launches Claude Fable 5.2 and Mythos 5.2 - OfficeChai