Fable5 Extended to July 12 with Cost-Saving Orchestration Architecture
新智元 · wechat · 2026-07-08
Anthropic has extended the free period for Fable5 to July 12 (the weekly 50% limit remains unchanged; exceeding it requires buying credits) and officially released two cost-saving architectures, with the core idea being 'large models direct, small models execute'.
In Advisor mode, Sonnet5 acts as the primary executor, bringing in Fable5 only for critical checkpoints, achieving about 92% of Fable5's solo score on SWE-bench Pro at roughly 63% of the cost. In Orchestrator mode, Fable5 acts as the commander to break down tasks and assign them to Sonnet5 sub-agents, achieving 96% performance on BrowseComp at 46% cost. The cookbook's real-world bill shows: for verifying ticket and reservation policies for the top 10 US national parks (cross-referencing 20 facts against nps.gov), the divided team took about $1.61 and 194 seconds, while solo Fable5 took about $4 and 608 seconds—making it 2.5x cheaper and 3x faster, with over 80% of the tokens pushed to low-rate workers. The accompanying advisor tools and documentation are now open, signaling the usage paradigm that 'top-tier models don't need to run at top-tier capacity all the time'.
Related event: Anthropic Extends Paid Access to Fable 5 Until July 12(27 posts)→
More from coding & agent
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- First-ever Three.js Conference lands in Paris, with a panel on AI-shortened design workflows — OdinLovis · 2026-09-11
- Data engineering, not agent frameworks, is the real bottleneck for enterprise AI agents — dhruv2038 · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- Investment Analyst Asks How to Build a Claude-Based Diligence Agent Stack — Careless_Tie2286 · 2026-09-11