The same capability that cost enterprise teams $25 per million output tokens in April now costs $50 — but that number is half what Claude Mythos Preview charged, and the underlying model is stronger.
Frontier reasoning was locked behind preview access until now
Serious agentic workloads — multi-step code execution, long-horizon research, complex document reasoning — required either paying Mythos Preview rates or accepting a capability ceiling. That ceiling just moved.
Two names, one model, one critical difference
Anthropic Releases Claude Fable 5 and Claude Mythos 5 and Claude Mythos 5 run on the same base model. What you get depends on which door you walk through: Claude Fable 5 ships with safety classifiers intact for general API access, while Claude Mythos 5 has some of those classifiers removed and remains in limited release. Both support a 1 million token context window and allow up to 128k output tokens per request, making them viable for tasks that would have required chunking or summarization workarounds before.
Researchers and engineers feel this first
- Software engineers running autonomous coding agents who need a model that holds context across an entire codebase without losing the thread
- Research analysts processing dense technical literature who need accurate synthesis across hundreds of pages in a single pass
- Enterprise AI teams evaluating provider costs who now have a benchmark-leading model at a price point that changes build-vs-buy math
The gap between what was available in limited preview and what ships to the general API just closed significantly.
Anthropic is racing to own the agentic tier before competitors stabilize
OpenAI’s o3 and Google’s Gemini 2.5 Pro both compete directly in the long-context reasoning space, and Anthropic’s decision to cut Mythos Preview pricing by more than half on a stronger model signals that this tier is becoming a commodity fight. If agentic capability stops being a differentiator, pricing and reliability become the only levers left.
What teams are already doing with it
- Run end-to-end software engineering tasks with full repo context loaded
- Process and synthesize scientific research documents exceeding 500 pages
- Build multi-step agents that plan, execute, and self-correct without human checkpoints
- Analyze vision inputs alongside long-form text in a single request
Input is priced at $10 per million tokens and output at $50 per million tokens — less than half the cost of Claude Mythos Preview at launch.
Claude Fable 5 carries the same 1M context window as its limited-release counterpart, which means teams who assumed they needed Mythos 5 access may find Fable 5 covers their actual use case without the waitlist.
The main tradeoff: Claude Mythos 5, the version with lifted classifiers suited for more sensitive research or red-teaming use cases, is still not publicly available.
For teams already on GPT-4o or Gemini 2.5 Pro, switching costs are real but the benchmark gap warrants a test. For teams already on older Claude Opus tiers, this is a direct upgrade path with no architectural change required.
The Mythos-class tier is where the enterprise AI race is actually being run
The arrival of a publicly available Mythos-class model at this price point shifts what enterprise teams should expect as a baseline from their AI providers. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.