Anthropic launches Claude Fable 5.1 and Mythos 5.1, cutting cache-read pricing 75% but raising net per-task cost ~20% due to higher output token usage

Anthropic released two new flagship models, Claude Fable 5.1 and Claude Mythos 5.1, for coding and knowledge work. Base input, output and cache-write pricing stayed the same ($10/$50/$12.5 per million tokens), while cache-read pricing dropped 75% (from $1.00 to $0.25 per MTok). Benchmark scores rose substantially (Terminal-Bench-Science 24.7% to 52.6%, HLE 55.5% to 59.1%), but because the models generate roughly 1.7x more output tokens per task, the net cost per task actually increased by about 20% compared with Fable 5.

Developers and enterprises budgeting for Claude-based agentic or long-context workflows need to recalculate costs rather than assume the cache-price cut lowers their bills, since higher output volume likely outweighs the cache savings for typical tasks. This is a useful, concrete warning against taking headline pricing cuts at face value. Competitive impact is modest: this reshuffles benchmark rankings among Anthropic, OpenAI and xAI's existing frontier models rather than shifting market power to a new player.

The "new SOTA" and "reclaiming the frontier lead" framing glosses over the fact that overall task cost went up, not down, which undercuts the pricing announcement as a customer win. The claim that Fable 5.1 and Mythos 5.1 share identical underlying weights, differentiated only by safety classifier routing, is explicitly unconfirmed community speculation and should not be treated as fact. No primary source links (official Anthropic pricing page, benchmark methodology, or model card) were captured, so all figures trace back to a newsletter aggregating tweets and third-party Artificial Analysis measurements.

1. 01Whether Anthropic publishes an official model card or benchmark methodology confirming the AA Intelligence Index 66 figure, rather than relying on Artificial Analysis's third-party measurement
2. 02Independent developer cost reports over the next 4-6 weeks testing whether the ~20% net cost increase per task holds across real agentic workloads, not just Anthropic's own benchmark tasks
3. 03Whether Anthropic confirms or denies the community claim that Fable 5.1 and Mythos 5.1 share identical weights differentiated only by safety routing, which would materially change how developers choose between them