Updated October 10, 2026. Rank says Claude Opus 4.8 is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
Both of these models will change. Get the price, version and retirement notices for the pair, each with its source. Follow model changes
Claude Opus 4.8 has the higher public score estimate, 69.12 versus 67.79, but the 90% score intervals overlap. Treat that as a lead, not a settled winner. 27 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
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.
Code generation, repair, and software-engineering tasks
GPT-5.5
GPT-5.5 has the higher public coding point estimate, 62.1 to 60.8, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
Tool use, computer use, and multi-step task completion
Claude Opus 4.8
Claude Opus 4.8 has the higher public agentic point estimate, 61.6 to 60.7, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
1K fresh input + 500 output tokens
Claude Opus 4.8
Claude Opus 4.8 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
GPT-5.5
GPT-5.5 has the lower estimated token cost for this stated workload. Claude Opus 4.8 has no published cached-input rate, so cached tokens use its listed input rate.
50K fresh input + 3K output tokens
Claude Opus 4.8
Claude Opus 4.8 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Prompts that approach the documented context limit
No clear pick
The documented context windows are equal.
Choose a job from the LLM Selector's task catalog to see the category row it rests on, under the same basis rules as the table below. A directional row stays directional; choosing a job never creates a winner.
The same task catalog as the LLM Selector. Each job names the evidence surface it rests on; nothing here adds a new score.
Like-for-like · BenchAlign v5.8
GPT-5.5 has the higher coding point estimate. Conditional score ranges do not establish rank confidence.
Coding scores combine specific tasks and setups. Match the editor, harness, and effort to your workflow.
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
2 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
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.
ARC-AGI-2Reasoning
Normalized gap 12.9OfficeQA ProMultimodal
Normalized gap 12.1SWE-bench ProCoding
Normalized gap 10.6HLE w/o toolsKnowledge
Normalized gap 8.4OSWorld 2.0Agentic
Normalized gap 7.6Each row shows the public-lane category score for both models: the BenchAlign v5.8 lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.
| Category | Claude Opus 4.8 | GPT-5.5 | Basis | Reading |
|---|---|---|---|---|
| Agentic | 61.6Supported · #20/123 | 60.7Supported · #22/123 | Like-for-likeBenchAlign v5.8 lane · 12 vs 14 public rows | Claude Opus 4.8 leads · intervals overlap |
| Coding | 60.8Supported · #19/146 | 62.1Supported · #15/146 | Like-for-likeBenchAlign v5.8 lane · 11 vs 10 public rows | GPT-5.5 leads · intervals overlap |
| Reasoning | 62.9#24/28 | 69.7#20/28 | Like-for-likeProvisional lane · 2 vs 2 weighted rows | GPT-5.5 leads |
| Knowledge | 70.7Supported · #14/177 | 70.6Supported · #15/177 | Like-for-likeBenchAlign v5.8 lane · 6 vs 6 public rows | Practical tie |
| Multimodal | 91.2#5/54 | 75.5#23/54 | Directional onlyProvisional lane · 2 vs 2 weighted rows | Directional only |
| Instruction following | 75.4#61/127 | 93.3#8/127 | Directional onlyProvisional lane · 0 vs 0 weighted rows | Directional only |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Math | 65.7#2/7 | 69.9Unranked · 3 rankable rows | Not comparableProvisional lane · 3 vs 2 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign v5.8) differ from the provisional-lane categories. Unranked scores sit on the provisional lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
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.
1K fresh input + 500 output tokens
Claude Opus 4.8 has the lower modeled cost
Costs use the listed standard API rates.
50K fresh input + 3K output tokens
Claude Opus 4.8 has the lower modeled cost
Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
GPT-5.5 has the lower modeled cost
Claude Opus 4.8 has no published cached-input rate, so cached tokens use its listed input rate.
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
Maximum documented context; output-token limits may be lower.
Claude Opus 4.8
GPT-5.5
Claude Opus 4.8
claude-opus-4-8
Anthropic model overviewGPT-5.5
gpt-5.5
OpenAI pricingA missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude Opus 4.8
Not published
GPT-5.5
$0.5 per 1M cached input tokens
OpenAI pricingClaude Opus 4.8
text, image
Anthropic model overviewGPT-5.5
Not sourced
Claude Opus 4.8
GPT-5.5
Not sourced
Claude Opus 4.8
Generally Available · Claude API
Anthropic model overviewGPT-5.5
Not sourced
Claude Opus 4.8
Reasoning
GPT-5.5
Reasoning
Claude Opus 4.8
Proprietary
GPT-5.5
Proprietary
Claude Opus 4.8
Proprietary
GPT-5.5
Proprietary
Claude Opus 4.8
2026-05-28
GPT-5.5
2026-04-23
Claude Opus 4.8 has the higher public score estimate, 69.12 versus 67.79, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
GPT-5.5 has the higher public coding point estimate, 62.1 to 60.8, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
Claude Opus 4.8 has the higher public agentic tasks point estimate, 61.6 to 60.7, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
For the stated presets, chat costs $0.0175 on Claude Opus 4.8 and $0.02 on GPT-5.5; repository review costs $0.325 and $0.34; the cache-heavy agent loop costs $1.35 and $0.5. Claude Opus 4.8 has no published cached-input rate, so cached tokens use its listed input rate.
Both models list the same context window, 1M.
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Terminal-Bench 3.0
Not directly comparable
Terminal-Bench 2.1
Not directly comparable
BrowseComp
GPT-5.5 leads this result
DeepSearchQA
Not directly comparable
OSWorld-Verified
Claude Opus 4.8 leads this result
Finance Agent v2
Not directly comparable
MCP Atlas
Claude Opus 4.8 leads this result
Toolathlon
Claude Opus 4.8 leads this result
Gert Labs
Shared sourceClaude Opus 4.8 leads this result
ResearchClawBench
Shared sourceClaude Opus 4.8 leads this result
OSWorld 2.0
Shared sourceClaude Opus 4.8 leads this result
Terminal-Bench 2.1 (Vals)
GPT-5.5 leads this result
Terminal-Bench 2.0
Not directly comparable
CyberGym
Not directly comparable
τ²-bench results
Not directly comparable
JobBench
Not directly comparable
ExploitGym
Not directly comparable
ApprenticeBench
Not directly comparable
SWE-bench Verified
Not directly comparable
SWE-bench Pro
Claude Opus 4.8 leads this result
SWE Multilingual
Not directly comparable
SWE Multimodal
Not directly comparable
Terminal-Bench 2.1
Not directly comparable
cursorBench31
Shared sourceGPT-5.5 leads this result
CursorBench 3.2
Shared sourceClaude Opus 4.8 leads this result
FrontierCode 1.1 Main
Shared sourceClaude Opus 4.8 leads this result
LiveCodeBench (Vals)
Claude Opus 4.8 leads this result
SWE-bench (Vals)
Claude Opus 4.8 leads this result
PostTrainBench v1.1
Shared sourceClaude Opus 4.8 leads this result
Terminal-Bench 2.0
Not directly comparable
Vibe Code Bench
Not directly comparable
React Native Evals
Not directly comparable
ARC-AGI-2
GPT-5.5 leads this result
ARC-AGI-3
Shared sourceClaude Opus 4.8 leads this result
ARC-AGI-1
Shared sourceGPT-5.5 leads this result
MRCR v2 64K-128K
Not directly comparable
MRCR v2 128K-256K
Not directly comparable
OfficeQA Pro
Claude Opus 4.8 leads this result
ScreenSpot Pro
Not directly comparable
CharXiv
Not directly comparable
CharXiv w/o tools
Not directly comparable
MMMU-Pro
Not directly comparable
MMMU-Pro w/ Python
Not directly comparable
GPQA
Tie
GPQA-D
Tie
HLE
Claude Opus 4.8 leads this result
HLE w/o tools
Claude Opus 4.8 leads this result
GPQA Diamond (Vals)
GPT-5.5 leads this result
MMLU-Pro (Vals)
Claude Opus 4.8 leads this result
INCLUDE
Not directly comparable
USAMO 2026
Not directly comparable
FrontierMath v2 (Tiers 1-3)
Shared sourceGPT-5.5 leads this result
FrontierMath v2 (Tier 4)
Shared sourceGPT-5.5 leads this result
FrontierMath (legacy)
Not directly comparable
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 5,500+ readers.
Last updated October 10, 2026
