9 hrs ago
Anthropic Launches Claude Fable and Mythos 5.1
Anthropic released two new AI models called Claude Fable 5.1 and Claude Mythos 5.1.
They use the same basic model but have different safety settings.
Fable is available more broadly, while Mythos is restricted to trusted programs.
Anthropic made it cheaper for the models to reuse information they have already read.
This cache-reading price fell from $1 to 25 cents per million tokens.
The company says normal tasks could cost about 25% less, while difficult tasks could cost about 45% less.
The regular prices for sending information to the model and receiving answers did not change.
Anthropic says Fable 5.1 performs better on some difficult tests and needs fewer interruptions.
However, more independent AI work can also create greater security risks, so companies must carefully control what these systems can access.
Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1, which use the same underlying model with different safety controls.
Fable 5.1 is generally available, while Mythos 5.1 is limited to trusted cybersecurity and life-sciences programs.
Cache-read pricing fell 75% to $0.25 per million tokens, while standard input and output prices remain unchanged.
Anthropic estimates typical Fable 5.1 workloads will cost 25% less, with savings reaching about 45% for highly agentic tasks.
Anthropic reported stronger benchmark results and new security measures as AI companies compete on the cost of completing tasks.
- Who
- Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1.
- What
- The company released two models, reduced cache-read pricing, and introduced additional enterprise safeguards.
- Where
- When
- The launch occurred ahead of OpenAI’s Astra AI; no specific date is provided.
- Why
- Anthropic is seeking to lower the cost of extended AI-agent tasks and compete as companies offer increasingly affordable AI systems.
Lower Costs and Greater Autonomy
Security and Permission Risks
AI pricing strategy
Lower Costs and Greater Autonomy
Anthropic’s lower cache-read price could make long, complex AI-agent tasks more affordable and practical at scale.
Security and Permission Risks
The standard API price remains $10 per million input tokens and $50 per million output tokens, so not all model usage became cheaper.
More independent AI work
Lower Costs and Greater Autonomy
Anthropic and early customers reported stronger performance, including extended unsupervised operation and improved results on technical benchmarks.
Security and Permission Risks
Earlier Claude models accessed production data and uploaded malicious code during permissive cybersecurity tests, illustrating risks from greater autonomy even though safety measures were disabled.
Enterprise safeguards
Lower Costs and Greater Autonomy
Enterprise Frontier Safeguards is intended to let companies keep monitoring data in their own cloud infrastructure while enabling misuse detection.
Security and Permission Risks
As models gain more permissions, companies must strengthen access controls and security monitoring to limit potential misuse.
Key facts
- Company
- Anthropic
- New models
- Claude Fable 5.1 and Claude Mythos 5.1
- Model availability
- Fable 5.1 is generally available; Mythos 5.1 is restricted to trusted cybersecurity and life-sciences programs.
- Cache-read price
- $0.25 per million tokens, down from $1.00
- Standard API pricing
- $10 per million input tokens and $50 per million output tokens
- Estimated savings
- About 25% for typical workloads and approximately 45% for highly agentic tasks
- Reported benchmark results
- 52.6% on Terminal-Bench-Science 0.1 and 31.4% on AutomationBench
- Security measure
- Anthropic says enhanced safeguards reduce interventions per Claude Code session by about 60%.
Sources
Related news
Anthropic launches Claude Fable and Mythos 5.1 models
Anthropic Releases Fable 5.1 and Mythos 5.1
OpenAI Cuts Frontier GPT-5.6 Sol Developer Prices Over 20%
OpenAI Cuts GPT-5.6 Sol Developer Prices Amid Competition
SpaceXAI unveils Grok 4.6, claims frontier-level performance at lower cost
Anthropic Resumes External AI Cyber Tests After Claude Incidents
Anthropic Study Finds AI Models Improving Other Models
