Reasoning
Like-for-like- Claude Opus 5
- 90.4
- GPT-5.6 Sol
- 92.5
- Weighted basis
- 1 vs 1 rows
- Reading
- GPT-5.6 Sol leads
Claude Opus 5 vs GPT-5.6 Sol
Updated August 22, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Claude Opus 5 has the higher public score estimate, 82.72 versus 81.73, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
11 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.
Which one for your work
Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.
Long documents
Prompts that approach the documented context limit
GPT-5.6 Sol
GPT-5.6 Sol has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Claude Opus 5
Claude Opus 5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Claude Opus 5
Claude Opus 5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Show secondary and unsupported calls
Repository review cost
50K fresh input + 3K output tokens
Claude Opus 5
Claude Opus 5 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
52 Claude Opus 5 only
11 Shared
14 GPT-5.6 Sol only
- Shared results
- 11
- Claude Opus 5 only
- 52
- GPT-5.6 Sol only
- 14
- Like-for-like categories
- 1 / 8
2 categories use different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.
Agentic
Directional only- Claude Opus 5
- 90.8
- GPT-5.6 Sol
- 92.0
- Weighted basis
- 1 vs 2 rows
- Reading
- Directional only
Coding
Directional only- Claude Opus 5
- 89.5
- GPT-5.6 Sol
- 64.6
- Weighted basis
- 2 vs 1 rows
- Reading
- Directional only
Knowledge
Not comparable- Claude Opus 5
- 64.7
- GPT-5.6 Sol
- 94.6
- Weighted basis
- 1 vs 1 rows
- Reading
- Not comparable
Math
Not comparable- Claude Opus 5
- Not measured
- GPT-5.6 Sol
- 87.5
- Weighted basis
- 0 vs 2 rows
- Reading
- Not comparable
Multilingual
Not comparable- Claude Opus 5
- Not measured
- GPT-5.6 Sol
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Multimodal
Not comparable- Claude Opus 5
- 66.9
- GPT-5.6 Sol
- 83.0
- Weighted basis
- 1 vs 1 rows
- Reading
- Not comparable
Instruction following
Not comparable- Claude Opus 5
- Not measured
- GPT-5.6 Sol
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
| Category | Claude Opus 5 | GPT-5.6 Sol | Weighted basis | Reading |
|---|---|---|---|---|
| Reasoning | 90.4 | 92.5 | Like-for-like1 vs 1 rows | GPT-5.6 Sol leads |
| Agentic | 90.8 | 92.0 | Directional only1 vs 2 rows | Directional only |
| Coding | 89.5 | 64.6 | Directional only2 vs 1 rows | Directional only |
| Knowledge | 64.7 | 94.6 | Not comparable1 vs 1 rows | Not comparable |
| Math | Not measured | 87.5 | Not comparable0 vs 2 rows | Not comparable |
| Multilingual | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Multimodal | 66.9 | 83.0 | Not comparable1 vs 1 rows | Not comparable |
| Instruction following | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
Shape of the matched evidence
Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.
Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.
SWE-bench Pro
Coding
ARC-AGI-2
Reasoning
BrowseComp
Agentic
What each workload costs
Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.
Chat turn
1K fresh input + 500 output tokens
- Claude Opus 5
- $0.0175
- Fits in one request
- GPT-5.6 Sol
- $0.02
- Fits in one request
Claude Opus 5 has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- Claude Opus 5
- $0.325
- Fits in one request
- GPT-5.6 Sol
- $0.34
- Fits in one request
Claude Opus 5 has the lower modeled cost
Costs use the listed standard API rates.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Claude Opus 5
- $0.45
- Fits in one request
- GPT-5.6 Sol
- $0.5
- Fits in one request
Claude Opus 5 has the lower modeled cost
Costs use the listed standard API rates.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationClaude Opus 5GPT-5.6 Sol
Context window
Maximum documented context; output-token limits may be lower.
Claude Opus 5
GPT-5.6 Sol
1.05M
OpenAI model catalogAPI model ID
Claude Opus 5
claude-opus-5
Anthropic model overviewGPT-5.6 Sol
gpt-5.6-sol
OpenAI model catalogCached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude Opus 5
$0.5 per 1M cached input tokens
Claude API pricingGPT-5.6 Sol
$0.5 per 1M cached input tokens
OpenAI pricingDocumented inputs
Claude Opus 5
text, image
Anthropic model overviewGPT-5.6 Sol
text, image
OpenAI model catalogDocumented outputs
Claude Opus 5
GPT-5.6 Sol
Provider availability
Claude Opus 5
Generally Available · Claude API
Anthropic model overviewGPT-5.6 Sol
Generally Available · OpenAI Responses API
OpenAI model catalogReasoning profile
Claude Opus 5
Reasoning
GPT-5.6 Sol
Reasoning
Weight access
Claude Opus 5
Proprietary
GPT-5.6 Sol
Proprietary
License
Claude Opus 5
Proprietary
GPT-5.6 Sol
Proprietary
Release date
Claude Opus 5
2026-07-24
GPT-5.6 Sol
2026-07-09
If you already use one of these models
- Deployment change
- The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
- Quality signal
- Claude Opus 5 has the higher public score estimate, 82.72 versus 81.73, but the 90% score intervals overlap.
- Workload cost
- Repository review: $0.325 vs $0.34. Cache-heavy agent loop: $0.45 vs $0.5.
- Context tradeoff
- GPT-5.6 Sol has the larger documented window (1.05M).
Run the same representative tasks against both endpoints before changing production traffic.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence77 rows
Agentic
Terminal-Bench 3.0
Shared sourceClaude Opus 542.7%GPT-5.6 Sol34.6%Claude Opus 5 leads this result
BrowseComp
Claude Opus 590.8%SourceGPT-5.6 Sol92.2%SourceGPT-5.6 Sol leads this result
HLE w/ tools
Claude Opus 564.7%SourceGPT-5.6 Sol—Not directly comparable
DeepSearchQA
Claude Opus 595.0%SourceGPT-5.6 Sol—Not directly comparable
DRACO
Claude Opus 588.6%SourceGPT-5.6 Sol—Not directly comparable
BrowseComp (10-agent, prerelease)
Claude Opus 593.6%SourceGPT-5.6 Sol—Not directly comparable
OSWorld 2.0
Claude Opus 570.6%SourceGPT-5.6 Sol62.6%SourceClaude Opus 5 leads this result
MCP Atlas
Claude Opus 585.8%SourceGPT-5.6 Sol—Not directly comparable
MCP-Atlas claim coverage
Claude Opus 589.1%SourceGPT-5.6 Sol—Not directly comparable
LAB all-pass (Anthropic harness)
Claude Opus 523.58%SourceGPT-5.6 Sol—Not directly comparable
LAB criterion-pass (Anthropic harness)
Claude Opus 593.74%SourceGPT-5.6 Sol—Not directly comparable
LAB all-pass (Harvey held-out)
Claude Opus 511.7%SourceGPT-5.6 Sol—Not directly comparable
LAB criterion-pass (Harvey held-out)
Claude Opus 594.1%SourceGPT-5.6 Sol—Not directly comparable
Toolathlon-Verified
Claude Opus 580.6%SourceGPT-5.6 Sol—Not directly comparable
Toolathlon Verified Pass@3
Claude Opus 587.0%SourceGPT-5.6 Sol—Not directly comparable
Toolathlon Verified Pass³
Claude Opus 573.1%SourceGPT-5.6 Sol—Not directly comparable
Toolathlon Verified avg. turns
Claude Opus 523.5 turnsSourceGPT-5.6 Sol—Not directly comparable
AutomationBench
Claude Opus 526.0%SourceGPT-5.6 Sol—Not directly comparable
Terminal-Bench 2.0
Claude Opus 5—GPT-5.6 Sol91.9%SourceNot directly comparable
CyberGym
Claude Opus 5—GPT-5.6 Sol84.5%SourceNot directly comparable
ExploitGym
Claude Opus 5—GPT-5.6 Sol33.7%SourceNot directly comparable
Toolathlon
Claude Opus 5—GPT-5.6 Sol58%SourceNot directly comparable
Coding
SWE-bench Verified
Claude Opus 596%SourceGPT-5.6 Sol—Not directly comparable
SWE-bench Pro
Claude Opus 579.2%SourceGPT-5.6 Sol64.6%SourceClaude Opus 5 leads this result
SWE Multilingual
Claude Opus 589.5%SourceGPT-5.6 Sol—Not directly comparable
SWE Multimodal
Claude Opus 559.4%SourceGPT-5.6 Sol—Not directly comparable
deepSwe
Claude Opus 568.8%SourceGPT-5.6 Sol72.7%SourceGPT-5.6 Sol leads this result
FrontierCode 1.1 Main
Claude Opus 553.4%SourceGPT-5.6 Sol—Not directly comparable
FrontierCode 1.1 Extended
Claude Opus 563.6%SourceGPT-5.6 Sol60.6%SourceClaude Opus 5 leads this result
ProgramBench (episode 1)
Claude Opus 583.0%SourceGPT-5.6 Sol—Not directly comparable
ProgramBench
Claude Opus 593.0%SourceGPT-5.6 Sol—Not directly comparable
cursorBench32
Shared sourceClaude Opus 570.0%GPT-5.6 Sol67.2%Claude Opus 5 leads this result
VulcanBench v3
Claude Opus 587.0%SourceGPT-5.6 Sol87.0%SourceTie
Terminal-Bench 2.0
Claude Opus 5—GPT-5.6 Sol91.9%SourceNot directly comparable
Reasoning
ARC-AGI-1
Claude Opus 597.50%SourceGPT-5.6 Sol—Not directly comparable
ARC-AGI-2
Claude Opus 590.4%SourceGPT-5.6 Sol92.5%SourceGPT-5.6 Sol leads this result
ARC-AGI-3
Claude Opus 530.2%SourceGPT-5.6 Sol7.8%SourceClaude Opus 5 leads this result
GeneBench-Pro
Claude Opus 5—GPT-5.6 Sol28.7%SourceNot directly comparable
Knowledge
HLE
Claude Opus 564.7%SourceGPT-5.6 Sol—Not directly comparable
HLE w/o tools
Claude Opus 556.3%SourceGPT-5.6 Sol—Not directly comparable
HealthBench (raw)
Claude Opus 567.1%SourceGPT-5.6 Sol—Not directly comparable
HealthBench (length-adjusted)
Claude Opus 557.8%SourceGPT-5.6 Sol—Not directly comparable
HealthBench Professional
Claude Opus 559.8%SourceGPT-5.6 Sol60.5%SourceGPT-5.6 Sol leads this result
HealthBench Professional (raw)
Claude Opus 573.4%SourceGPT-5.6 Sol—Not directly comparable
BioMysteryBench (human-solvable)
Claude Opus 590.1%SourceGPT-5.6 Sol—Not directly comparable
BioMysteryBench (human-difficult)
Claude Opus 549.4%SourceGPT-5.6 Sol—Not directly comparable
SpatialBench Verified
Claude Opus 572.5%SourceGPT-5.6 Sol—Not directly comparable
SingleCellBench
Claude Opus 560.6%SourceGPT-5.6 Sol—Not directly comparable
ProteinGym Hard
Claude Opus 547.7%SourceGPT-5.6 Sol—Not directly comparable
Protein Design
Claude Opus 542.5%SourceGPT-5.6 Sol—Not directly comparable
Organic chemistry V2
Claude Opus 561.6%SourceGPT-5.6 Sol—Not directly comparable
Protocols (troubleshooting)
Claude Opus 561.1%SourceGPT-5.6 Sol—Not directly comparable
Protocols (understanding)
Claude Opus 578.4%SourceGPT-5.6 Sol—Not directly comparable
GPQA
Claude Opus 5—GPT-5.6 Sol94.6%SourceNot directly comparable
GPQA-D
Claude Opus 5—GPT-5.6 Sol94.6%SourceNot directly comparable
HealthBench Hard
Claude Opus 5—GPT-5.6 Sol33.1%SourceNot directly comparable
Math
IMO 2026
Claude Opus 542/42SourceGPT-5.6 Sol—Not directly comparable
RiemannBench (no tools)
Claude Opus 560.0%SourceGPT-5.6 Sol—Not directly comparable
RiemannBench (tools)
Claude Opus 579.0%SourceGPT-5.6 Sol—Not directly comparable
ArXivMath Jun. 2026 (no tools)
Claude Opus 590.8%SourceGPT-5.6 Sol—Not directly comparable
ArXivMath Jun. 2026 (tools)
Claude Opus 591.3%SourceGPT-5.6 Sol—Not directly comparable
FrontierMath (legacy)
Claude Opus 5—GPT-5.6 Sol89%SourceNot directly comparable
FrontierMath v2 (Tiers 1-3)
Claude Opus 5—GPT-5.6 Sol89.000%SourceNot directly comparable
FrontierMath v2 (Tier 4)
Claude Opus 5—GPT-5.6 Sol83.000%SourceNot directly comparable
Multilingual
Multimodal
Chartography (no tools)
Claude Opus 529.6%SourceGPT-5.6 Sol—Not directly comparable
Chartography (tools)
Claude Opus 583.0%SourceGPT-5.6 Sol—Not directly comparable
BenchCAD Vision2Code (no tools)
Claude Opus 50.366SourceGPT-5.6 Sol—Not directly comparable
BenchCAD Vision2Code (tools)
Claude Opus 50.821SourceGPT-5.6 Sol—Not directly comparable
GDP.pdf (no tools)
Claude Opus 583.4%SourceGPT-5.6 Sol—Not directly comparable
GDP.pdf (tools)
Claude Opus 585.5%SourceGPT-5.6 Sol—Not directly comparable
OfficeQA
Claude Opus 578.1%SourceGPT-5.6 Sol—Not directly comparable
OfficeQA Pro
Claude Opus 566.9%SourceGPT-5.6 Sol—Not directly comparable
MMMU-Pro
Claude Opus 5—GPT-5.6 Sol83%SourceNot directly comparable
MMMU-Pro w/ Python
Claude Opus 5—GPT-5.6 Sol84.6%SourceNot directly comparable
Frequently asked questions
Which is better, Claude Opus 5 or GPT-5.6 Sol?
Claude Opus 5 has the higher public score estimate, 82.72 versus 81.73, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Which is better for coding, Claude Opus 5 or GPT-5.6 Sol?
The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, Claude Opus 5 or GPT-5.6 Sol?
The current agentic tasks averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which costs less, Claude Opus 5 or GPT-5.6 Sol?
For the stated presets, chat costs $0.0175 on Claude Opus 5 and $0.02 on GPT-5.6 Sol; repository review costs $0.325 and $0.34; the cache-heavy agent loop costs $0.45 and $0.5. Costs use the listed standard API rates.
Which has the larger context window, Claude Opus 5 or GPT-5.6 Sol?
GPT-5.6 Sol has the larger documented context window: 1.05M, compared with 1M.
Related comparisons
Last updated August 22, 2026
Watch Claude Opus 5 vs GPT-5.6 Sol
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.