Fable 5
A competing model positioned against GPT-5.6 Sol. On the Agent Ultimate Exam benchmark, it scores 13.1 points lower than GPT-5.6 Sol.
Core Facts
Timeline (last 90 days)
On the quality-cost Pareto frontier, DeepSeek V4.1 Flash sits at the far upper left, beating models such as Opus 5, Fable 5, GPT 6.1 Soul, Gemini 3.8 Flash, and GLM 5.3 Max.
GVS5H claims that five Qwen3.8-27B instances working in coordination match Fable 5 on LiveCodeBench Hard.
Until the configuration of the GVS5H Harness and the details of the comparison baseline are made public, the claim that five Qwen3.8-27B match Fable 5 should be treated as a hypothesis awaiting verification rather than an established fact.
Fable 5 is not commonly seen on mainstream public leaderboards, and the original post did not explain what it specifically refers to.
A post appeared on Hacker News whose title claims that five Qwen3.8-27B models match Fable 5 with the help of the GVS5H Harness, and the post received 60 upvotes and 15 comments.
When Fable 5 and Opus 4.6 were released, the author felt unsure how to use them.
Simon Willison's analysis suggests that the internal enablement of the Fable and Mythos series models was the key catalytic factor in Anthropic's leap in coding speed.
Anthropic's three steep jumps in merge volume correspond respectively to the releases of Fable 5 and Mythos 5 (June 9), Opus 5 (July 24), and Fable 5.1 and Mythos 5.1 (September 1).
The version orchestrated by Fable 5, which only provided guidance without modifying code, achieved more realistic physics and control logic; Fable itself changed only one or two lines of code, with nearly all code written by Quen, and the final product quality was noticeably better.
When orchestrator Fable provided instructions containing an error, the Quen small model did not follow them blindly but instead implemented the correct behavior based on how real instruments actually work.
40 more timeline events
All Facts (20)
The API model identifier for GPT-6 Astra is gpt-6-astra.
90%VerifiedAccounting for real usage scenarios, the actual cost advantage of Opus 5 over Fable is approximately 20% to 25%, with Fable costing about $2.75 per task and Opus 5 about $2.03.
65%UnverifiedFable 5.1 has stronger coding capabilities.
60%UnverifiedGrok 4.6's score on the Deep Suite benchmark improved from 54% to 66%.
60%UnverifiedOn the quality-cost Pareto frontier, DeepSeek V4.1 Flash sits at the far upper left, beating models such as Opus 5, Fable 5, GPT 6.1 Soul, Gemini 3.8 Flash, and GLM 5.3 Max.
50%UnverifiedGVS5H claims that five Qwen3.8-27B instances working in coordination match Fable 5 on LiveCodeBench Hard.
50%UnverifiedUntil the configuration of the GVS5H Harness and the details of the comparison baseline are made public, the claim that five Qwen3.8-27B match Fable 5 should be treated as a hypothesis awaiting verification rather than an established fact.
50%UnverifiedFable 5 is not commonly seen on mainstream public leaderboards, and the original post did not explain what it specifically refers to.
50%UnverifiedWhen Fable 5 and Opus 4.6 were released, the author felt unsure how to use them.
50%UnverifiedSimon Willison's analysis suggests that the internal enablement of the Fable and Mythos series models was the key catalytic factor in Anthropic's leap in coding speed.
50%UnverifiedThe version orchestrated by Fable 5, which only provided guidance without modifying code, achieved more realistic physics and control logic; Fable itself changed only one or two lines of code, with nearly all code written by Quen, and the final product quality was noticeably better.
50%UnverifiedDuring the orchestration experiment, the entire build process had approximately 30 substantive defects, which were resolved across four rounds of fixes with none remaining in the final version, all of which were discovered through external validation and never proactively detected by the model itself.
50%UnverifiedSol scored 68.8% in the coding test, close to the previous-generation Fable 5's 69.9%, but with a per-task cost approximately 80% lower.
50%UnverifiedLuna scored 66.6% in the coding test, with costs 93% and 96% lower than Opus 5 and Fable 5, respectively.
50%UnverifiedThe video author judges Grok 4.7 to be roughly close to the level of Fable 5, but not quite reaching Fable 5.1.
50%Unverified"Fable 5" is not a common name in Anthropic's publicly known product line.
50%UnverifiedThe entire project ultimately consumed approximately 2 billion tokens, with Fable 5 and Opus 5 each using about 800 million, Codex close to 400 million, and GROK using 76 million.
50%UnverifiedFable 5 is an AI product designed for creative writing or role-playing scenarios.
50%Unverified"Fable 5" does not correspond to any currently mainstream and known large model product.
50%UnverifiedIn the task of a robotic arm grasping blocks into a bowl, Fable 5 had a success rate of 5%, Fable 5.1 reached 40%, and GPT-6 Astra achieved 95%.
50%