What changed in Claude 5.2
Anthropic says the models share an underlying foundation. Fable includes safeguards for broad availability, while organizations seeking Mythos for cybersecurity or biomedical research must apply through verification programs.
API speed and pricing
The company reports generation up to 30% faster than the 5.1 models. Standard pricing is $8 per million input tokens and $40 per million output tokens. Optional fast mode costs $16 and $80 respectively, with claimed generation speedups of up to 2.5 times.
Evaluate completed tasks, not impressive answers
For a development team, a useful comparison measures completed work: whether a change passes tests, whether the requirements were understood and how much human correction it needs. A single migration mistake can outweigh time saved on many quick requests.
A practical evaluation can start with a small collection of previously solved problems. Give each model equivalent context, establish the expected outcomes and apply the same scoring criteria. Include cases where the correct response is to acknowledge missing evidence rather than produce a confident explanation.
Token prices are only part of the cost
Our recommendation is to measure the complete cycle: requests, retries, verification and specialist time. Two short calls can cost more in practice than one long call when the output needs extensive manual repair. Cost per accepted result is a more useful internal metric than cost per request alone.
Test fast mode where waiting actually blocks work. Overnight batch processing and an interactive assistant can involve different tradeoffs. Evaluate Ukrainian language performance and your own terminology explicitly rather than treating a broad benchmark ranking as an answer for every workload.
Before moving a production workflow
Our assessment: the update is most relevant to teams with difficult, measurable tasks. Before switching, define an acceptable cost per successful operation, an error threshold and which data may be sent to the service. A limited pilot should provide more actionable evidence than replacing every workflow at once.
The performance figures above are company claims, not our independent test results. Choosing between Fable and Mythos also requires understanding the relevant access conditions rather than simply selecting the more specialized name.
