Anthropic went CRAZY (Mythos/Fable 5.1)

New Anthropic frontier model with the same per-token pricing as Fable 5 but a 75% cache-read price cut, claimed ~25% cheaper for typical workloads. Benchmarks improved (Terminal Bench Science, Cursor Bench, computer use, GDPval). Artificial Analysis found it scores 66 but costs more per task because it uses 1.7x output tokens. Creator's tests (landscape scenes, Rubik's cube, five websites, slides) were good but behind GPT 5.6 Soul.

EFEnterprise Frontier SafeguardsAnthropicUnverified

Zero data retention option where data is stored in customer-controlled cloud infrastructure, though Anthropic can still read it for misuse detection. The creator sees it as a half measure that may not satisfy enterprises.

HEhere.nowUnverified

Skill that lets agents publish web pages, reports, games, PDFs and slides to a shareable link. It is free to use. The creator used it to publish the Fable 5.1 tests.

Same underlying model as Fable 5.1 with different guardrails. It scores higher than Fable 5.1 on Terminal Bench at every reasoning effort, up to about 5% better at max, and is slightly cheaper. Anthropic says it reward hacks less than Mythos 5.

GSGPT-5.6 SolOpenAILimited in EU

Used as a comparison. It scores 61 on Artificial Analysis at about 43 cents per task, and the creator calls it the best bargain. Its output on the creator's landscape test was better than Fable 5.1's.

G5GLM 5.3Z.aiAvailable in EU

Cheap model at 68 cents per task on Artificial Analysis. Its test outputs were very similar to Fable 5.1's, which the creator suspects is distillation.