Across 49 comparable benchmarks, Claude Opus 5 has better reported results on 44 and worse on 4, with 1 tied. 5 further shared results are listed below, each with the reason it is not counted.
Moderate evidence49 of 54 shared results countedHow we compare
Price basis: Listed rates vs standard rates. The price section below shows the applicable periods.
| Fact | GPT-5.5 | Claude Opus 5 |
|---|---|---|
| Context window | 1M tokens | 1M tokens |
| Input / 1M | $5.00 | $5.00 |
| Output / 1M | $30.00 | $25.00 |
| Blended (4:1) | $6.76 | $5.76 |
| Status | Available | Available |
| Released | April 23, 2026 | July 24, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates โ set your own mix.
Mean of 33 shared benchmarks grouped by capability on a 0โ100 scale. How averages are built
Claude Opus 5 ahead on 12 of 12 counted
Claude Opus 5 ahead on 3 of 3 counted
Claude Opus 5 ahead on 3 of 4 counted
Claude Opus 5 ahead on 4 of 5 counted
Claude Opus 5 ahead on 4 of 4 counted
Claude Opus 5 ahead on 1 of 1 counted
Claude Opus 5 ahead on 2 of 2 counted
Split 1โ1 of 2 counted
Claude Opus 5 ahead on 10 of 11 counted
No shared comparable results for Cybersecurity.
Published scores, with matching modes where available. Gaps show GPT-5.5 relative to Claude Opus 5. Reading the results
| Benchmark | GPT-5.5 | Claude Opus 5 | Gap |
|---|---|---|---|
| APEX-SWERepository Engineering ยท Pass@1 | 37.0Extra-High ยท tools | 63.7โ Max ยท tools | โ26.7 pts#38 vs #3 of 61 |
| Code MigrationMaintenance & Optimization ยท Hidden-test pass rate (%) | 45.2Extra-High ยท tools | 57.5โ Max ยท tools | โ12.3 pts#19 vs #9 of 79 |
| DeepSWERepository Engineering ยท Pass@1 (DeepSWE v1.1) | 67.0Extra-High ยท tools | 73.7โ Max ยท tools | โ6.6 pts#36 vs #16 of 73 |
| FrontierCode 1.1Code Generation & Editing ยท Weighted rubric score (%) | 56.7Extra-High ยท toolsยท frontiercode-1-1-extended | 63.6โ Unspecified Reasoning ยท toolsยท frontiercode-1-1-extended | โ6.9 pts#8 vs #3 of 51 |
| FrontierCode 2.0Repository Engineering ยท Weighted rubric score | 22.6Extra-High ยท tools + internet | 26.7Extra-High ยท tools + internet | 0.85ร#20 vs #15 of 36 |
| PostTrain BenchML Engineering ยท Score | 27.2Extra-High ยท tools | 35.0โ Reported best | 0.78ร#12 vs #8 of 15 |
| SWE-Atlas-QnARepository Engineering ยท Task Resolve Rate (%) | 45.4Extra-High ยท tools | 63.2Extra-High ยท tools | โ17.7 pts#13 vs #2 of 17 |
| SWE-Bench Pro - PublicRepository Engineering ยท Accuracy | 58.6Thinking High ยท tools | 79.2โ Max ยท tools | โ20.6 pts#22 vs #6 of 66 |
| Terminal-Bench 2.1Agentic Development ยท Accuracy | 83.1Extra-High ยท tools | 89.1โ Max ยท tools | โ6.0 pts#33 vs #10 of 68 |
| Terminal-Bench 4.0Agentic Development ยท Resolution rate (%) | 14.6Extra-High ยท tools | 53.9Extra-High ยท tools | โ39.3 pts#54 vs #16 of 90 |
| Vibe Code Bench v1.1Code Generation & Editing ยท Pass rate (%) | 69.8Extra-High ยท tools | 88.4โ Max ยท tools | โ18.6 pts#51 vs #13 of 121 |
| WeirdML v2ML Engineering ยท Average accuracy across 17 tasks (%) | 84.9Extra-High ยท tools | 91.6โ Thinking High ยท tools | โ6.7 pts#4 vs #1 of 39 |
| ARC-AGI-1Abstract Generalization ยท Score (%) | 95.0Extra-High ยท no tools | 97.5โ Thinking High ยท no tools | โ2.5 pts#20 vs #10 of 87 |
| ARC-AGI-2Abstract Generalization ยท Score (% solved); cost per task (USD)Not counted ยท Speed and cost telemetry is reported, not scored | 85.0Extra-High ยท no tools | 90.4โ Max ยท no tools | 0.94ร#14 vs #9 of 80 |
| ARC-AGI-3Abstract Generalization | 0.43Thinking High ยท no toolsยท arc3-semi-private-standard | 30.2Thinking High ยท no toolsยท arc3-semi-private-standard | 0.01ร#19 vs #7 of 26 |
| SimpleBenchCommonsense ยท Score (AVG@5) | 69.0Normal ยท no tools | 80.6โ Thinking High ยท no tools | 0.86ร#31 vs #14 of 106 |
| Vending-Bench 2Logical Planning ยท Money Balance (USD)Not counted ยท Speed and cost telemetry is reported, not scored | 7524Unspecified Reasoning ยท tools | 11182Unspecified Reasoning ยท tools | 0.67ร#22 vs #9 of 64 |
| FrontierMath ErdลsMathematics ยท Accuracy (% of 68 open Erdลs problems resolved) | 0.0Extra-High ยท tools | 0.0โ Max ยท tools | =#7 vs #7 of 14 |
| FrontierMath Tier 4 v2Mathematics ยท Accuracy (verification_code) | 72.5Extra-High ยท tools | 73.2โ Max ยท tools | โ0.7 pts#16 vs #15 of 73 |
| FrontierMath v2Mathematics ยท Accuracy (verification_code) | 85.3Extra-High ยท tools | 85.6โ Max ยท tools | โ0.4 pts#16 vs #15 of 85 |
| Riemann-benchMathematics ยท Accuracy (%) | 55.2Extra-High ยท tools + internet | 68.0โ Max ยท tools + web | โ12.8 pts#10 vs #4 of 42 |
| CritPtScientific Reasoning ยท Score | 27.1Extra-High ยท no tools | 29.1โ Max ยท no tools | 0.93ร#18 vs #13 of 134 |
| GPQA DiamondScientific Reasoning ยท Accuracy | 94.0Extra-High ยท no tools | 93.7โ Thinking High ยท no tools | +0.3 pts#19 vs #24 of 200 |
| SciCodeScientific Computing ยท Score | 56.1Thinking High ยท no tools | 56.4โ Max ยท no tools | 0.99ร#28 vs #27 of 98 |
| MedCodeClinical Workflows ยท Accuracy (%) | 49.1Extra-High ยท tools | 63.6โ Max ยท tools | โ14.5 pts#26 vs #1 of 97 |
| MedScribeClinical Workflows ยท Rubric pass rate (%) | 86.9Extra-High ยท tools | 91.0โ Max ยท tools | โ4.1 pts#20 vs #4 of 99 |
| Agents' Last Exam v1.0Tool Orchestration ยท Pass rate | 26.6Extra-High ยท toolsยท pass-rate | 32.2โ Thinking High ยท toolsยท pass-rate | โ5.6 pts#17 vs #6 of 34 |
| APEX-AgentsTool Orchestration ยท Score | 55.1Reported best | 65.8โ Max ยท tools | 0.84ร#27 vs #9 of 61 |
| MCP-AtlasTool Orchestration ยท Pass rate / claim coverage | 75.3Extra-High ยท tools | 85.8Extra-High ยท tools | โ10.5 pts#32 vs #5 of 55 |
| OSWorld 2.0Desktop Workflows ยท Partial scoreNot counted ยท Different or unrecorded test environments โ excluded from the tally | 49.5Extra-High ยท toolsยท osworld-full-108 | 70.2โ Max ยท toolsยท osworld-offline-82 | โ#13 vs #3 of 19 |
| OSWorld-VerifiedDesktop Workflows ยท Accuracy | 78.7Thinking High ยท tools | 83.4โ Unspecified Reasoning ยท tools | โ4.7 pts#13 vs #5 of 36 |
| BrowseCompFact Finding ยท Accuracy | 84.4Thinking High ยท tools + web | 90.8โ Max ยท tools + web | โ6.4 pts#14 vs #4 of 61 |
| Context ArenaMemory & Persistence ยท Accuracy (8 needles, 4K-128K context) | 94.2Extra-High ยท no tools | 97.7โ Max ยท no tools | โ3.5 pts#5 vs #1 of 61 |
| EBR-benchMemory & Persistence ยท Score (% of 21 objectives, best of the two scored playthroughs) | 34.3Extra-High ยท tools | 45.7โ Max ยท tools | โ11.4 pts#14 vs #10 of 30 |
| GDP.pdfDocuments & Charts ยท Task pass rate (%) | 22.6Thinking High ยท no tools | 21.6โ Max ยท no tools | +1.0 pts#26 vs #28 of 96 |
| MMMU-ProVisual Understanding ยท Accuracy | 81.2Thinking Medium ยท no tools | 84.7โ Max ยท no tools | โ3.5 pts#26 vs #10 of 136 |
| Creative WritingWriting | 1845Normal ยท no tools | 2135Normal ยท no tools | 0.86ร#17 vs #4 of 121 |
| AA-AnalystAgentData Analysis ยท pass^5 (%) | 50.0Deep Thinking ยท tools + web | 53.8โ Max ยท tools + web | โ3.8 pts#11 vs #7 of 39 |
| AutomationBenchOffice & Business ยท Pass RateNot counted ยท Different or unrecorded test environments โ excluded from the tally | 16.6Extra-High ยท toolsยท zapier-private-heldout | 50.3โ Max ยท toolsยท legacy-default | โ#44 vs #14 of 64 |
| EMB (Excel Modeling)Finance ยท Accuracy (%) | 64.5Extra-High ยท tools | 73.6โ Max ยท tools | โ9.0 pts#28 vs #8 of 78 |
| Finance Agent v2Finance ยท Score | 51.8Extra-High ยท tools | 58.6โ Max ยท tools | 0.88ร#37 vs #9 of 75 |
| Harvey Lab-AALegal ยท Criterion pass rate (some entries report all-pass rate) | 86.3Extra-High ยท tools | 93.5โ Max ยท tools | โ7.2 pts#14 vs #4 of 39 |
| Harvey's Legal Agent BenchmarkLegal ยท Accuracy (%) | 3.8Extra-High ยท tools | 6.7โ Max ยท tools | โ2.9 pts#31 vs #21 of 52 |
| Legal Research BenchLegal ยท Score | 40.4Extra-High ยท tools | 55.3โ Max ยท tools | 0.73ร#36 vs #3 of 91 |
| Public Benefits Bench v1.1Service Workflows ยท Criterion pass rate (%) | 60.9Extra-High ยท tools | 76.9โ Max ยท tools | โ16.0 pts#32 vs #1 of 50 |
| SAGEService Workflows ยท Accuracy (%) | 51.5Extra-High ยท tools | 49.4โ Max ยท tools | +2.1 pts#13 vs #21 of 82 |
| Tax Agent BenchFinance ยท Accuracy (%, rubric ร citation quality) | 60.5Extra-High ยท tools | 75.1โ Max ยท tools | โ14.6 pts#18 vs #2 of 74 |
| ฯยณ-BankingService Workflows ยท Score | 44.6Extra-High ยท tools | 48.7โ Max ยท tools | 0.92ร#15 vs #6 of 120 |
| HLEKnowledge Exams ยท Accuracy | 52.2Thinking High ยท tools | 63.6โ Max ยท tools | โ11.4 pts#40 vs #8 of 151 |
| ECICapability Indices ยท ECI score (capability index, higher is better) | 159Reported best | 163Reported best | 0.98ร#10 vs #6 of 168 |
| LiveBenchCross-capability Suites ยท Accuracy | 80.2Extra-High ยท no tools | 80.1โ Max ยท no tools | +0.1 pts#10 vs #11 of 106 |
| LiveBench 2.0Cross-capability Suites ยท Overall score (equal-weight mean of 7 suite scores, 0-100) | 68.7Extra-High ยท tools | 82.1โ Max ยท task tools | 0.84ร#28 vs #8 of 51 |
| Text Arena (Coding)Preference Arenas ยท Arena Score | 1505Extra-High ยท no tools | 1712โ Max ยท no tools | 0.88ร#17 vs #1 of 28 |
| Vals IndexCapability IndicesNot counted ยท This metric has no better direction | 57.4Extra-High ยท tools | 67.2โ Max ยท tools | โ |
Repository Engineering ยท Pass@1
Maintenance & Optimization ยท Hidden-test pass rate (%)
Repository Engineering ยท Pass@1 (DeepSWE v1.1)
Code Generation & Editing ยท Weighted rubric score (%)
Repository Engineering ยท Weighted rubric score
ML Engineering ยท Score
Repository Engineering ยท Task Resolve Rate (%)
Repository Engineering ยท Accuracy
Agentic Development ยท Accuracy
Agentic Development ยท Resolution rate (%)
Code Generation & Editing ยท Pass rate (%)
ML Engineering ยท Average accuracy across 17 tasks (%)
Abstract Generalization ยท Score (%)
Abstract Generalization ยท Score (% solved); cost per task (USD)
Not counted ยท Speed and cost telemetry is reported, not scored
Abstract Generalization
Commonsense ยท Score (AVG@5)
Logical Planning ยท Money Balance (USD)
Not counted ยท Speed and cost telemetry is reported, not scored
Mathematics ยท Accuracy (% of 68 open Erdลs problems resolved)
Mathematics ยท Accuracy (verification_code)
Mathematics ยท Accuracy (verification_code)
Mathematics ยท Accuracy (%)
Scientific Reasoning ยท Score
Scientific Reasoning ยท Accuracy
Scientific Computing ยท Score
Clinical Workflows ยท Accuracy (%)
Clinical Workflows ยท Rubric pass rate (%)
Tool Orchestration ยท Pass rate
Tool Orchestration ยท Score
Tool Orchestration ยท Pass rate / claim coverage
Desktop Workflows ยท Partial score
Not counted ยท Different or unrecorded test environments โ excluded from the tally
Desktop Workflows ยท Accuracy
Fact Finding ยท Accuracy
Memory & Persistence ยท Accuracy (8 needles, 4K-128K context)
Memory & Persistence ยท Score (% of 21 objectives, best of the two scored playthroughs)
Writing
Data Analysis ยท pass^5 (%)
Office & Business ยท Pass Rate
Not counted ยท Different or unrecorded test environments โ excluded from the tally
Finance ยท Accuracy (%)
Finance ยท Score
Legal ยท Criterion pass rate (some entries report all-pass rate)
Legal ยท Accuracy (%)
Legal ยท Score
Service Workflows ยท Criterion pass rate (%)
Service Workflows ยท Accuracy (%)
Finance ยท Accuracy (%, rubric ร citation quality)
Service Workflows ยท Score
Knowledge Exams ยท Accuracy
Capability Indices ยท ECI score (capability index, higher is better)
Cross-capability Suites ยท Accuracy
Cross-capability Suites ยท Overall score (equal-weight mean of 7 suite scores, 0-100)
Preference Arenas ยท Arena Score
Capability Indices
Not counted ยท This metric has no better direction
A dash means no result is available.
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix ยท 5% steps
Blended price = effective input rate at 90% cache hits ร 80% + output rate ร 20%
GPT-5.5 costs17%morethan Claude Opus 5 per million tokens.Using listed rates vs standard rates at this token mix.
GPT-5.5: Standard pricing. context_length <= 272000 OpenAI official pricing (developers.openai.com/api/docs/pricing) verified 2026-10-21; standard tier; <=272K input tokens context_length > 272000 OpenAI official pricing (developers.openai.com/api/docs/pricing) verified 2026-10-21; standard tier; >272K input tokens; start date not published, verification date used
Claude Opus 5: Standard pricing. Anthropic official pricing (platform.claude.com/docs/en/about-claude/pricing) verified 2026-10-21; full 1M context at standard price
| Price / 1M tokens | GPT-5.5 | Claude Opus 5 | Ratio |
|---|---|---|---|
| Input | $5.00 | $5.00 | 1.00ร |
| Output | $30.00 | $25.00 | 1.20ร |
| Cached input | $0.50 | $0.50 | 1.00ร |
| Cache write | โ | โ | โ |
| Blended | $6.76 | $5.76 | 1.17ร |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
GPT-5.5 relative to Claude Opus 5
Summary: 7 of 8 shared benchmarks have matching recorded settings and sources. The average is a geometric mean of 7 positive-cost ratios, one equal weight per benchmark. This benchmark set can include related tests and indexes; it is not a workload cost estimate.
GPT-5.5 4.3ร cheaper
GPT-5.5 2.3ร cheaper
GPT-5.5 2.2ร cheaper
GPT-5.5 1.2ร more expensive
GPT-5.5 1.2ร cheaper
GPT-5.5 1.1ร cheaper
Settings or sources differ
GPT-5.5 4.3ร cheaper
GPT-5.5 2.3ร cheaper
GPT-5.5 2.2ร cheaper
GPT-5.5 1.2ร more expensive
GPT-5.5 1.2ร cheaper
GPT-5.5 1.1ร cheaper
Settings or sources differ
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Effort, protocol, aggregation and source match; other run conditions may differ.
GPT-5.5
Claude Opus 5
Settings or sources differ or are not recorded; interpret the comparison with care.
GPT-5.5
Claude Opus 5
| Benchmark | GPT-5.5 | Claude Opus 5 |
|---|---|---|
| AA Intelligence Index v4.3 | Not reported | 2 more settings |
| AA Intelligence Index v5.0.1 | Not reported | |
| AA-LCR | Not reported | |
| AA-Omniscience | Not reported | 2 more settings |
| Agent Arena | Not reported | $4.6032 / task Median ยท Net improvement index (percentage points vs. average model): 7.1 Max (With Tools + Internet) |
| Agent Arena Index | Not reported | $4.6109 / task Median ยท Agent Arena Index rating (index points): 1176 Max (With Tools + Internet) |
| AIME 2026 | Not reported | |
| CursorBench 4.0 | Not reported | $11.95 / task Average ยท Task score: 46.6 Max (With Tools) ยท cursorbench-4-0-cursor-agent $9.00 / task Average ยท Task score: 44.7 Thinking High (With Tools) ยท cursorbench-4-0-cursor-agent $4.87 / task Average ยท Task score: 40.7 Thinking Low (With Tools) ยท cursorbench-4-0-cursor-agent 2 more settings$6.94 / task Average ยท Task score: 43.3 Thinking Medium (With Tools) ยท cursorbench-4-0-cursor-agent $11.43 / task Average ยท Task score: 46.1 Extra-High (With Tools) ยท cursorbench-4-0-cursor-agent |
| FrontierSWE v2 | Not reported | |
| HLE | Not reported | $0.0625 / task Average ยท Accuracy: 43.4 Thinking Low (No Tools) ยท Text only ยท text-only $0.187 / task Average ยท Accuracy: 51.3 Thinking Medium (No Tools) ยท Text only ยท text-only 2 more settings$0.3521 / task Average ยท Accuracy: 52.8 Thinking High (No Tools) ยท Text only ยท text-only $0.5171 / task Average ยท Accuracy: 54.4 Extra-High (No Tools) ยท Text only ยท text-only |
| SWE-sweep | Not reported | |
| Tax Agent Bench | Not reported | $5.1366 / task Average ยท Accuracy (%, rubric ร citation quality): 45.5 Unspecified Reasoning (With Tools) |
| Terminal-Bench 4.0 | Not reported | 3 more settings$18.60 / task Average ยท Resolution rate (%): 53.5 Unspecified Reasoning (With Tools) ยท vals-mini-swe |
| Terminal-Bench 4.1 | Not reported | 2 more settings |
| Vals Index v2.1 | Not reported | |
| Vibe Code Bench 1-100 | Not reported | |
| WeirdML v3 | Not reported | $25.2622 / task Average ยท Official score (0-1; higher is better): 0.19 Extra-High (With Tools) |
| Attribute | GPT-5.5 | Claude Opus 5 |
|---|---|---|
| Availability | Available | Available |
| Provider | OpenAI | Anthropic |
| Model type | Reasoning model | Reasoning model |
| First release | April 23, 2026 | July 24, 2026 |
| Context window | 1M tokens | 1M tokens |
| Max output | 128K tokens | 128K tokens |
| Architecture | Undisclosed | Undisclosed |
| Documented capabilities | Reasoning, Multilingual, Coding, Tool use, Function calling, Computer use, Web search, Long context | Reasoning, Multilingual, Coding, Tool use, Function calling, Computer use, Long context |
| Knowledge cutoff | โ | 2026-05 |
| Input modalities | Text, Image | Text, Image |
| Output modalities | Text | Text |
| Weights | Closed / API only | Closed / API only |
| Commercial use | Proprietary | Proprietary |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used GPT-5.5 or Claude Opus 5? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.