Updated October 17, 2026. Rank says GPT-5.4 is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
GPT-5.4 has the higher public score estimate, 66.7 versus 55.89, but the 90% score intervals overlap. Treat that as a lead, not a settled winner. 16 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-5.4
GPT-5.4 has the higher public coding point estimate, 47.7 to 44.7, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
Tool use, computer use, and multi-step task completion
GPT-5.4
GPT-5.4 has the higher public agentic point estimate, 52.1 to 46.3, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
Prompts that approach the documented context limit
GPT-5.4
GPT-5.4 has the larger documented context window.
1K fresh input + 500 output tokens
GPT-5.4
GPT-5.4 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
50K fresh input + 3K output tokens
GPT-5.4
GPT-5.4 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
Not enough matched evidence
The page does not recommend a cost winner because at least one model cannot fit the stated workload in one request. Claude Sonnet 4.6 does not fit this workload in one request. Claude Sonnet 4.6 has no published cached-input rate, so cached tokens use its listed input rate.
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-5.4 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.
Terminal-Bench 2.0Agentic
Normalized gap 16.0ARC-AGI-2Reasoning
Normalized gap 15.6CharXivMultimodal
Normalized gap 5.4HLEKnowledge
Normalized gap 3.1OSWorld-VerifiedAgentic
Normalized gap 2.9Each 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 Sonnet 4.6 | GPT-5.4 | Basis | Reading |
|---|---|---|---|---|
| Agentic | 46.3Supported · #55/128 | 52.1Supported · #44/128 | Like-for-likeBenchAlign v5.8 lane · 10 vs 15 public rows | GPT-5.4 leads · intervals overlap |
| Coding | 44.7Supported · #54/150 | 47.7Supported · #48/150 | Like-for-likeBenchAlign v5.8 lane · 9 vs 6 public rows | GPT-5.4 leads · intervals overlap |
| Knowledge | 55.4Supported · #58/184 | 69.1Supported · #20/184 | Like-for-likeBenchAlign v5.8 lane · 6 vs 7 public rows | GPT-5.4 leads · intervals overlap |
| Multimodal | 58.7#45/54 | 73.1#30/54 | Directional onlyProvisional lane · 1 vs 3 weighted rows | Directional only |
| Instruction following | 48.0#99/148 | 90.7#23/148 | Directional onlyProvisional lane · 0 vs 0 weighted rows | Directional only |
| Reasoning | 62.8Unranked · 4 rankable rows | 64.2#23/28 | Not comparableProvisional lane · 1 vs 2 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Math | Not ranked | 82.0Unranked · 2 rankable rows | Not comparableProvisional lane · 0 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
GPT-5.4 has the lower modeled cost
Costs use the listed standard API rates.
50K fresh input + 3K output tokens
GPT-5.4 has the lower modeled cost
Costs use the listed standard API rates.
200K cached + 20K fresh input + 10K output tokens
Claude Sonnet 4.6 does not fit this workload in one request. Claude Sonnet 4.6 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 Sonnet 4.6
200K
GPT-5.4
1.05M
OpenAI pricingClaude Sonnet 4.6
Not sourced
GPT-5.4
gpt-5.4
OpenAI pricingA missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude Sonnet 4.6
Not published
GPT-5.4
$0.25 per 1M cached input tokens
OpenAI pricingClaude Sonnet 4.6
Not sourced
GPT-5.4
Not sourced
Claude Sonnet 4.6
Not sourced
GPT-5.4
Not sourced
Claude Sonnet 4.6
Not sourced
GPT-5.4
Not sourced
Claude Sonnet 4.6
Non-Reasoning
GPT-5.4
Reasoning
Claude Sonnet 4.6
Proprietary
GPT-5.4
Proprietary
Claude Sonnet 4.6
Proprietary
GPT-5.4
Proprietary
Claude Sonnet 4.6
2026-02-01
GPT-5.4
2026-03-05
GPT-5.4 has the higher public score estimate, 66.7 versus 55.89, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
GPT-5.4 has the higher public coding point estimate, 47.7 to 44.7, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
GPT-5.4 has the higher public agentic tasks point estimate, 52.1 to 46.3, with Supported evidence for both models. The conditional ranges do not establish rank confidence.
For the stated presets, chat costs $0.0105 on Claude Sonnet 4.6 and $0.01 on GPT-5.4; repository review costs $0.195 and $0.17; the cache-heavy agent loop costs $0.81 and $0.25. Claude Sonnet 4.6 does not fit this workload in one request. Claude Sonnet 4.6 has no published cached-input rate, so cached tokens use its listed input rate.
GPT-5.4 has the larger documented context window: 1.05M, compared with 200K.
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Terminal-Bench 2.0
GPT-5.4 leads this result
OSWorld-Verified
GPT-5.4 leads this result
Claw-Eval
Shared sourceClaude Sonnet 4.6 leads this result
CyberGym
GPT-5.4 leads this result
Gert Labs
Shared sourceGPT-5.4 leads this result
OSWorld 2.0
Not directly comparable
JobBench
Shared sourceGPT-5.4 leads this result
Terminal-Bench 2.1 (Vals)
Not directly comparable
ApprenticeBench
Shared sourceGPT-5.4 leads this result
MazeBench
Tie
BrowseComp
Not directly comparable
MCP Atlas
Not directly comparable
Toolathlon
Not directly comparable
τ²-bench results
Not directly comparable
DeepSearchQA
Not directly comparable
ResearchClawBench
Not directly comparable
ExploitGym
Not directly comparable
SWE-bench Verified
Not directly comparable
SWE-Rebench
Not directly comparable
React Native Evals
Shared sourceGPT-5.4 leads this result
Terminal-Bench 2.0
GPT-5.4 leads this result
Vibe Code Bench
Shared sourceGPT-5.4 leads this result
cursorBench31
Not directly comparable
FrontierCode 1.1 Main
Not directly comparable
LiveCodeBench (Vals)
Not directly comparable
SWE-bench (Vals)
Not directly comparable
LiveCodeBench Pro
Not directly comparable
SWE-bench Pro
Not directly comparable
PostTrainBench v1.1
Not directly comparable
ARC-AGI-2
Shared sourceGPT-5.4 leads this result
ARC-AGI-1
Shared sourceGPT-5.4 leads this result
ARC-AGI-3
Not directly comparable
CharXiv
GPT-5.4 leads this result
MMMU-Pro
Not directly comparable
OfficeQA Pro
Not directly comparable
MMMU-Pro w/ Python
Not directly comparable
ERQA
Not directly comparable
SimpleVQA
Not directly comparable
ScreenSpot Pro
Not directly comparable
ZeroBench
Not directly comparable
MedXpertQA (MM)
Not directly comparable
GPQA
GPT-5.4 leads this result
SuperGPQA
Not directly comparable
MMLU-Pro
Not directly comparable
HLE
GPT-5.4 leads this result
GPQA Diamond (Vals)
Not directly comparable
MMLU-Pro (Vals)
Not directly comparable
HLE w/o tools
Not directly comparable
GPQA-D
Not directly comparable
HealthBench Hard
Not directly comparable
MedXpertQA (Text)
Not directly comparable
HealthBench Professional
Not directly comparable
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 5,600+ readers.
Last updated October 17, 2026
