Fable 5.1 Tops Benchmarks but Sparks Cost Controversy
After Fable Studio (Anthropic) released Fable 5.1, in-depth benchmarks from Artificial Analysis and extensive user testing show the model topping multiple benchmarks—but with significantly higher running costs and token consumption at high effort tiers, sparking widespread debate over whether it's "smarter or just pricier." What can currently be confirmed is that all three of the following hold simultaneously: leading performance, a pricier max tier, and strong cost-efficiency at the low-effort tier.
Confirmed
- Artificial Analysis benchmarks: Fable 5.1 (max) scored 66 on the Intelligence Index, surpassing Opus 5 and GPT-5.6 Sol; cost per task was $3.76, 20% higher than Fable 5 (max) at $3.14; cheaper cached reads saved Anthropic roughly $1.40.
- By AA's total running cost, Fable 5.1 (especially max reasoning mode) is 56% more expensive than Fable 5; token consumption on AAII tasks surged 73.5%, from about 83 million to about 144 million.
- Output tokens differ by 11x across the five effort tiers: about 13.1M at low, about 143.7M at max.
- CursorBench 3.2 (@haider1): Fable 5.1 Medium scored 68.0% at $3.53 per task, beating GPT-5.6 Sol Max's 67.2% at only about 60% of the cost; @kimmonismus also reported High beating Sol 5.6 Max while being cheaper.
- Fable 5.1 tops Senior SWE-Bench, with Medium as the best tier, roughly matching Fable 5 at about 50% of the cost.
- Low-cost usage paths: Anthropic engineer @RLanceMartin and @every both noted that low-effort mode matches high-effort on CursorBench at about 1/3 the cost, and is often competitive with Opus/Sonnet under per-task billing; @danielmac8 added that Low beats Mythos 5 High on Terminal-Bench 4.0.
Unconfirmed
- @tetsuoai reported a single prompt burning through 20x quota in 52 minutes without returning code; @robleclerc relayed that a user exhausted 5 hours of Claude quota in under 30 minutes (similar reports from both 5x and 20x speed users). These are individual reports—whether widespread or related to runaway agent counts remains unclear; @tetsuoai cautions users to watch how many Agents the model spins up.
- A screenshot from @WonderFactory showing $3.69 per task, higher than Fable 5, coexists with claims of "halved costs"; the discrepancy may stem from different tiers and cache configurations.
Why it matters
- The controversy hinges on two perspectives: @Angaisb and @scaling01 point out that while the absolute cost of frontier models is rising, the "cost per unit of intelligence" is falling—Fable 5.1 can halve per-task cost while matching or exceeding the previous intelligence index; some users on @m14 also question Anthropic's continued price hikes, contrasting them with OpenAI's price cuts.
- @VraserX observed that 5.1 burns through Pro quota faster than Opus 5, speculating this reflects significantly greater compute demands or capabilities; @robleclerc argues the model is too costly as a full-time team member and only fits part-time-style usage. For quota-based subscribers, tier selection and agent management will directly determine the experience and the bill.
2026-09-02 ~ 2026-09-02 · 21 related posts
- Episode 1: Claude Fable 5.1 Spotted on Bedrock and in Docs Ahead of Imminent Launch(2026-09-01, 5 posts)
- Episode 2: Anthropic Releases Claude Fable 5.1 and Mythos 5.1 with Big Bench Gains and 75% Cache Price Cut(2026-09-02, 70 posts)
- Episode 3: Fable 5.1 Tops Benchmarks but Sparks Cost Controversy(2026-09-02, 21 posts)
- Episode 4: Every's Hands-on: Anthropic Fable 5.1 Reclaims the Coding Crown at Half the Cost(2026-09-02, 8 posts)
Primary sources
- Claude Fable 5.1 tops benchmarks but costs 20% more per task due to higher token usage — ArtificialAnlys ·
- Fable 5.1 costs $3.76 per task; Anthropic's cache price cut saves $1.40 — ArtificialAnlys ·
- Tips for Claude Fable 5.1: Low-Effort Mode and Cost Optimization — RLanceMartin ·
- Claude Fable 5.1 Launches: High Usage Hints at Major Compute Jump — VraserX · 2026-09-02
- [source] Tips for Claude Fable 5.1: Low-Effort Mode and Cost Optimization — RLanceMartin · 2026-09-02
- Fable 5.1 Beats GPT-5.6 on Benchmark at Lower Cost — haider1 · 2026-09-02
- Hands-on review: Claude Fable 5.1 matches Opus at lower cost with prompt cache savings — every · 2026-09-02
- Fable 5.1 leaps ahead on price-performance, beating Sol 5.6 Max on Cursor Bench for less — kimmonismus · 2026-09-02
- Fable 5.1: Matches High-End Rivals at Lower Cost — daniel_mac8 · 2026-09-02
- Fable 5.1 Cost Found Higher Than Fable 5 in Real-World Test — WonderFactory · 2026-09-02
- Fable 5.1 is 56% more expensive to run than Fable 5 — Angaisb_ · 2026-09-02
- Fable 5.1 token usage surges 73.5% on AAII benchmark — Angaisb_ · 2026-09-02
- Fable 5.1 Max Reasoning Costs 56% More Than Version 5 — Angaisb_ · 2026-09-02
- Users Question Anthropic's Pricing Strategy vs OpenAI's Deflation — Angaisb_ · 2026-09-02
- Fable 5.1 Matches Performance of Fable 5 at Half the Cost — scaling01 · 2026-09-02
- Fable 5.1 Benchmarks Show Gains Over Sol 5.6 Max — eyishazyer · 2026-09-02
- [source] Claude Fable 5.1 tops benchmarks but costs 20% more per task due to higher token usage — ArtificialAnlys · 2026-09-02
- [source] Fable 5.1 costs $3.76 per task; Anthropic's cache price cut saves $1.40 — ArtificialAnlys · 2026-09-02
- Claude Fable 5.1's five effort levels span 11x in token usage, 58-66 on AA Index — ArtificialAnlys · 2026-09-02
- Fable 5.1 tops SWE-Bench, matching prior perf at 50% cost — ajratner · 2026-09-02
- Fable 5.1 drains Claude 5-hour limit in under 30 minutes, users report — robleclerc · 2026-09-02
- New Claude Model Burns Through Usage: 20x Limit in 52 Minutes via Single Prompt — tetsuoai · 2026-09-02