Claude Opus 4.6 vs Claude Opus 5.5
Different efforts: high · medium · Compare all at high
AutomationBench-AA AutomationBench-AA: Agents orchestrate SaaS workflows through APIs; objectives receive credit only when the task has no guardrail violation
Opus 4.6—
Opus 5.561.2
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)
Opus 4.667
Opus 5.584.3
DeepSWE 1.1 DeepSWE 1.1: Agents change repository code to resolve software engineering tasks under the stated DeepSWE edition
Opus 4.6 · max31.9
Opus 5.5 · max72.3
EBR Bench EBR Bench: Agents learn to play Earthborne Rangers across repeated playthroughs
Opus 4.6 · max12.7
Opus 5.5 · max71.4
Furniture Assembly Furniture Assembly: Models inspect furniture assembly photos, identify incorrect builds and locate the mistaken step
Opus 4.6 · max28.3
Opus 5.5 · max83.3
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
Opus 4.6—
Opus 5.554.2
MCP-Atlas Public MCPAtlas Public: MCPAtlas Public as reported in this launch evaluation; the post is the authority for the recorded task and grading scope
Opus 4.6 · max79
Opus 5.5—
MCP-Atlas · full 1000 MCP-Atlas · full 1000: Tool orchestration across MCP servers on the stated task split
Opus 4.6 · max76.8
Opus 5.5—
τ³-bench · banking · 1.0.1 τ-bench official · 3-banking-1.0.1: Customer-service agents converse with simulated users, retrieve domain knowledge and change state through tools
Opus 4.6 · max27.3
Opus 5.5—
AA Analyst Agent AA Analyst Agent: Data-analyst tasks: the model explores data with tools and reports findings
Opus 4.6—
Opus 5.5 · max56.3
ITBench SRE ITBench SRE: IT incident tasks: diagnose and fix broken systems like a site-reliability engineer
Opus 4.6—
Opus 5.5 · max38.2
APEX-Agents APEX-Agents: Agents complete long-horizon professional-services tasks across realistic applications
Opus 4.6 · max33
Opus 5.5—
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
Opus 4.6—
Opus 5.552.5
Terminal-Bench Hard (version unstated) Terminal-Bench Hard (version unstated): Agents solve difficult software, system-administration and data-processing tasks in a terminal
Opus 4.648.5
Opus 5.5—
Terminal-Bench Science 0.1 Terminal-Bench Science 0.1: Agents complete expert-authored research workflows across five scientific domains in a terminal
Opus 4.6—
Opus 5.543.3
AA-Briefcase 1.1 · Elo AA-Briefcase 1.1 · Elo: Agents produce professional deliverables; blind pairwise comparisons produce the source Elo rating
Opus 4.6—
Opus 5.51,625.1
FrontierCode 1.1 FrontierCode 1.1: Agents implement changes in real repositories, graded against maintainer-defined requirements
Opus 4.6—
Opus 5.557.1
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
Opus 4.6—
Opus 5.5 · max4.2
SWE-bench Pro (private) SWE-bench Pro (private): Agents resolve long-horizon professional repository tasks; public and private splits are separate
Opus 4.647.1
Opus 5.5—
SpireClimb SpireClimb: Agents play Slay the Spire under the named game and difficulty setting
Opus 4.6—
Opus 5.5 · max3.3
SpireClimb StS1 A20 Heart SpireClimb StS1 A20 Heart: Agents play Slay the Spire under the named game and difficulty setting
Opus 4.6—
Opus 5.5 · max0
SpireClimb StS2 A10 SpireClimb StS2 A10: Agents play Slay the Spire under the named game and difficulty setting
Opus 4.6—
Opus 5.5 · max6.7
| Benchmark | Opus 4.6 | Opus 5.5 |
|---|---|---|
| AutomationBench-AA | — | 61.2 |
| AA-LCR (long context) 1.1 | 67 | 84.3 |
| DeepSWE 1.1 | 31.9 · max | 72.3 · max |
| EBR Bench | 12.7 · max | 71.4 · max |
| Furniture Assembly | 28.3 · max | 83.3 · max |
| GDPval-AA (work tasks) 2.1 | — | 54.2 |
| MCP-Atlas Public | 79 · max | — |
| MCP-Atlas · full 1000 | 76.8 · max | — |
| τ³-bench · banking · 1.0.1 | 27.3 · max | — |
| AA Analyst Agent | — | 56.3 · max |
| ITBench SRE | — | 38.2 · max |
| APEX-Agents | 33 · max | — |
| Terminal-Bench 4.0 | — | 52.5 |
| Terminal-Bench Hard (version unstated) | 48.5 | — |
| Terminal-Bench Science 0.1 | — | 43.3 |
| AA-Briefcase 1.1 · Elo | — | 1,625.1 |
| FrontierCode 1.1 | — | 57.1 |
| Harvey LAB-AA 1.1 | — | 4.2 · max |
| SWE-bench Pro (private) | 47.1 | — |
| SpireClimb | — | 3.3 · max |
| SpireClimb StS1 A20 Heart | — | 0 · max |
| SpireClimb StS2 A10 | — | 6.7 · max |
Specs
| Spec | Opus 4.6 | Opus 5.5 |
|---|---|---|
| Input, per 1M | $5 | $4 |
| Output, per 1M | $25 | $20 |
| Cost per task? | — | $1.34 |
| Speed, tok/s | 45 | 81 |
| Context | 1M | 1M |
| Weights | Closed | Closed |
| Released | Feb 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
Opus 4.6 → Opus 5.5
No forced tool use›
Forced tool_choice any or tool returns HTTP 400; remove the forced choice.
Thinking is always on›
Disabled thinking or a manual budget returns HTTP 400; omit thinking or use adaptive thinking.
Remove sampling parameters›
Non-default temperature, top_p or top_k returns HTTP 400; omit these parameters and use prompting.
No assistant prefill›
Assistant prefill returns HTTP 400; use structured outputs or system instructions, with system instructions on Bedrock.
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.
Return thinking blocks unmodified›
Edited, reordered or partially dropped thinking blocks return HTTP 400; echo assistant messages complete and unmodified with tool results.
Opus 5.5 → Opus 4.6
Nothing known.