Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026, describing them as the world's most advanced models for coding and knowledge work. ^[1] Anthropic stated that new API accounts opened on or after August 31 can no longer manually edit Claude's prior context in multi-turn conversations while preserving the transcript of Claude's prior thinking. ^[2] Anthropic reported that over 16 million Claude exchanges were traced to distillation campaigns run through roughly 24,000 fake accounts, with Moonshot AI accounting for 3.4 million exchanges. ^[3] On Terminal-Bench 4.0, Claude Fable 5.1 achieved 55.8%, surpassing Opus 5 at 52.3% and OpenAI's GPT-5.6 Sol at 37.3%. ^[4] Anthropic claimed on September 1, 2026 that Claude Fable 5.1 and Claude Mythos 5.1 are the 'world's most advanced models for coding and knowledge work.' ^[5] Anthropic said on September 1, 2026 that Claude Code users should see around 60 percent fewer cybersecurity false positives with Claude Fable 5.1's improved safeguards. ^[6] Anthropic said on September 1, 2026 that Claude Fable 5.1 costs approximately 25 percent less than Fable 5 for typical workloads and up to 45 percent for highly agentic work due to reduced cache-read pricing, while base pricing remains the same. ^[7] Anthropic stated that Fable 5.1 would cost an estimated 25% less than Fable 5 for typical token-billed workloads, with savings up to 45% for highly agentic work. ^[8] Anthropic will add a watermark to outputs of models released after August 2, 2026, per the EU AI Act Code of Practice, and is rolling out a watermark detection API in private preview. ^[9] Anthropic announced the release directly, positioning Fable 5.1 specifically around autonomous, multi-step work and Mythos 5.1 as the paired release for knowledge work. ^[10] Anthropic kept list pricing for Fable 5.1 at $10 per million tokens for input, $50 per million tokens for output, and $12.5 per million tokens for cache write. ^[11] Anthropic claims that Claude Mythos 5.1 reduces false positive security blocks by 60 percent compared to its predecessor while analyzing vulnerabilities without generating exploits. ^[12]