Claude Fable 5.1 vs Claude Sonnet 5.5

$2.98 / task

Wins 9 of 19 benchmarks

$0.88 / task

Wins 4 of 19 benchmarks

Different efforts: medium · high · Compare all at high

Only shared

AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation

Fable 5.154.7

Sonnet 5.559.4

GDPval-AA (work tasks) 2.1 GDPval-AA (work tasks) 2.1: Real work tasks from many professions (reports, spreadsheets, analyses), graded against expert work

Fable 5.152.3

Sonnet 5.552.5

AA Analyst Agent AA Analyst Agent: Data-analyst tasks: the model explores data with tools and reports findings

Fable 5.1 · max57.5

Sonnet 5.5 · max57.5

Terminal-Bench 4.0 Terminal-Bench 4.0 · source setting 2: Agents use a terminal to complete executable coding, system administration and research tasks in the 4.0 suite

Fable 5.144.9

Sonnet 5.543.9

Terminal-Bench Science 0.1 Terminal-Bench Science 0.1: Agents complete expert-authored research workflows across five scientific domains in a terminal

Fable 5.1 · max43.3

Sonnet 5.534.3

AA-Briefcase 1.1 · Elo AA-Briefcase 1.1 · Elo: Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating

Fable 5.11,527.7

Sonnet 5.51,635.8

AA-LCR (long context) 1.1 AA-LCR (long context) 1.1: Questions that need reading and connecting facts across very long documents (around 100K tokens)

Fable 5.184.7

Sonnet 5.578

FrontierCode 1.1 FrontierCode 1.1: Agents implement changes in real repositories, graded against maintainer-defined requirements

Fable 5.1 · max49

Sonnet 5.544.3

Harvey LAB-AA 1.1 Harvey LAB-AA 1.1: Legal agents work from case documents; every rubric criterion must pass and material hallucinations receive no credit

Fable 5.1 · max6.4

Sonnet 5.5 · max2.8

SWE-bench Pro (v2 full) SWE-bench Pro (v2 full): Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate

Fable 5.1 · high99.1

Sonnet 5.5 · xhigh95.8

SWE-bench Pro (v2 hard) SWE-bench Pro (v2 hard): Agents resolve repository issues under the network-locked v2 protocol; full and 51-task Hard results are separate

Fable 5.1 · high92.2

Sonnet 5.5 · xhigh70.6

Furniture Assembly Furniture Assembly: Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step

Fable 5.1 · max70

Sonnet 5.5 · max75

SpireClimb SpireClimb: Agents play Slay the Spire under the named game and difficulty setting

Fable 5.1 · max2

Sonnet 5.5 · max1.3

SpireClimb StS1 A20 Heart SpireClimb StS1 A20 Heart: Agents play Slay the Spire under the named game and difficulty setting

Fable 5.1 · max0

Sonnet 5.5 · max0

SpireClimb StS2 A10 SpireClimb StS2 A10: Agents play Slay the Spire under the named game and difficulty setting

Fable 5.1 · max4

Sonnet 5.5 · max2.7

τ³-Banking · AA τ³-Banking · AA: Agents retrieve banking knowledge and execute tools to resolve customer-support workflows

Fable 5.141

Sonnet 5.5—

ITBench SRE ITBench SRE: IT incident tasks: diagnose and fix broken systems like a site-reliability engineer

Fable 5.1 · max49.5

Sonnet 5.5—

DeepSWE 1.1 DeepSWE 1.1: Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition

Fable 5.1—

Sonnet 5.5 · max70.5

EBR Bench EBR Bench: Agents learn to play Earthborne Rangers across repeated playthroughs

Fable 5.1 · max57.1

Sonnet 5.5—

Benchmark	Fable 5.1	Sonnet 5.5
AutomationBench-AA	54.7	59.4
GDPval-AA (work tasks) 2.1	52.3	52.5
AA Analyst Agent	57.5 · max	57.5 · max
Terminal-Bench 4.0	44.9	43.9
Terminal-Bench Science 0.1	43.3 · max	34.3
AA-Briefcase 1.1 · Elo	1,527.7	1,635.8
AA-LCR (long context) 1.1	84.7	78
FrontierCode 1.1	49 · max	44.3
Harvey LAB-AA 1.1	6.4 · max	2.8 · max
SWE-bench Pro (v2 full)	99.1 · high	95.8 · xhigh
SWE-bench Pro (v2 hard)	92.2 · high	70.6 · xhigh
Furniture Assembly	70 · max	75 · max
SpireClimb	2 · max	1.3 · max
SpireClimb StS1 A20 Heart	0 · max	0 · max
SpireClimb StS2 A10	4 · max	2.7 · max
τ³-Banking · AA	41	—
ITBench SRE	49.5 · max	—
DeepSWE 1.1	—	70.5 · max
EBR Bench	57.1 · max	—

Specs

Spec	Fable 5.1	Sonnet 5.5
Input, per 1M	$10	$2
Output, per 1M	$50	$10
Cost per task?	$2.98	$0.88
Speed, tok/s	56	106
Context	1M	1M
Weights	Closed	Closed
Released	Sep 2026	Sep 2026
See images	✓	✓
Read PDFs	Partial	✓
Hear audio	—	—
Watch video	—	—
Call tools	✓	✓
Force a tool call	Partial	—
Return strict JSON	✓	✓
Search the web	?	✓
Run code	?	✓
Use a computer	?	✓

What breaks when you switch

Fable 5.1 → Sonnet 5.5

No forced tool use›

Forced tool_choice any or tool returns HTTP 400; remove the forced choice.

Replace disabled thinking›

Disabled thinking returns HTTP 400; use between_tools for the lowest thinking setting.

No manual thinking budgets›

Manual thinking budgets and extra between_tools fields return HTTP 400; remove manual budgets and send between_tools without display, budget_tokens or block_binding.

Remove sampling parameters›

Non-default temperature, top_p or top_k returns HTTP 400; remove these parameters.

No assistant prefill›

A prefilled final assistant turn returns HTTP 400; remove that prefill.

Old computer tool rejected · Claude API / Google Cloud›

The computer_20251124 tool returns HTTP 400 on the Claude API and Google Cloud; use computer_toolset_20260801 there.

Sonnet 5.5 → Fable 5.1

Replace forced tool_choice›

Forced tool_choice returns HTTP 400; use auto with an explicit instruction and strict tools, or JSON outputs.

Adaptive thinking is always on›

Disabled thinking or manual budget_tokens returns HTTP 400; omit thinking or use adaptive thinking.

No non-default sampling values›

Non-default temperature, top_p or top_k returns HTTP 400; omit non-default sampling values.

Requires 30-day data retention · Claude API›

A Claude API organization or workspace without 30-day retention receives HTTP 400; configure workspace retention or contact the Anthropic account team.

Thinking blocks bind to conversations›

Where conversation binding is enforced, replaying an invalidated block returns HTTP 400; use the documented drop_block control to continue without it.

No assistant prefill›

Assistant prefill returns HTTP 400; use system prompt instructions instead.