Anthropic released its top-tier Fable 5.1 and Mythos 5.1 models on September 1, 2026, positioning them for coding, knowledge work and scientific research. Both use the same underlying foundation model, but Mythos 5.1 applies less restrictive safety guardrails and is available only to vetted cybersecurity and life sciences institutions in the United States. Fable 5.1 is available worldwide, including South Korea. The launch adds benchmark gains beyond Fable 5, including scores of 52.6% on Terminal-Bench-Science 0.1, 1,853 points on GDPval-AA v2, 31.4% on AutomationBench, 73.4% on CursorBench 3.2.0 and 55.8% on Terminal-Bench 4.0. Mythos 5.1 scored 60.9% on Terminal-Bench 4.0 because Fable's safeguards restrict certain cybersecurity tasks, not because the underlying model is more capable. Fable 5.1 also ranked first on the Artificial Analysis Intelligence Index with a score of 66, ahead of more than 192 models. Input and output prices remain $10 and $50 per million tokens, while cache-read pricing fell 75% to $0.25 per million tokens. Anthropic expects typical usage costs to fall about 25%, and agent workloads with extensive context and tools to become up to 45% cheaper, although demanding tasks can still cost more because they generate substantially more output. The models were used for protein-binder design, Venus mapping and faster protein and genomics inference. The release comes as Anthropic prepares for a possible IPO, expands enterprise safeguards and limits the preservation of prior reasoning records for new API accounts.