Skip to main content

Model comparison

Claude 3.5 Sonnet vs GPT-5.4 Pro

Data verified

Head-to-head evidence from 2 shared benchmark results across 1 category. Overall scores shown here use the public BenchAlign v5 ranking lane.

47.74/100
Margin
13.1pts
winning →
60.89/100
1 category wins1 category wins

Public leaderboard positions: Claude 3.5 Sonnet #132 (Estimated); GPT-5.4 Pro #44 (Estimated). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. Claude 3.5 Sonnet and GPT-5.4 Pro share 2 comparable benchmark results. 2 of 8 categories are comparable. 2 results are unique to Claude 3.5 Sonnet; 10 to GPT-5.4 Pro.

Updated July 23, 2026
Shared results
2
Claude 3.5 Sonnet only
2
GPT-5.4 Pro only
10
Comparable categories
2 / 8

Pick GPT-5.4 Pro if you want the stronger benchmark profile. Claude 3.5 Sonnet only becomes the better choice if knowledge is the priority or you want the cheaper token bill.

Confidence note. This is a partial-evidence comparison with 2 shared benchmark results across 1 evidence category; 2 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

GPT-5.4 Pro is clearly ahead on the BenchAlign aggregate, 60.89 to 47.74. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

GPT-5.4 Pro's sharpest advantage is in mathematics, where it averages 46.9 against 1.6. The single biggest benchmark swing on the page is FrontierMath v2 (Tiers 1-3), 2.069% to 50.000%. Claude 3.5 Sonnet does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.

GPT-5.4 Pro is also the more expensive model on tokens at $30.00 input / $180.00 output per 1M tokens, versus $3.00 input / $15.00 output per 1M tokens for Claude 3.5 Sonnet. That is roughly 12.0x on output cost alone. GPT-5.4 Pro is the reasoning model in the pair, while Claude 3.5 Sonnet is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. GPT-5.4 Pro gives you the larger context window at 1.05M, compared with 200K for Claude 3.5 Sonnet.

Category breakdown

Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.

Category scores and score margins for Claude 3.5 Sonnet and GPT-5.4 Pro
CategoryClaude 3.5 SonnetΔGPT-5.4 Pro
MathClaude 3.5 Sonnet1.6Margin 45.3GPT-5.4 Pro46.9
KnowledgeClaude 3.5 Sonnet59.4Margin 0.7GPT-5.4 Pro58.7
AgenticClaude 3.5 SonnetNot measuredMarginNo overlapGPT-5.4 Pro89.3
CodingClaude 3.5 Sonnet49.0MarginNo overlapGPT-5.4 ProNot measured
ReasoningClaude 3.5 SonnetNot measuredMarginNo overlapGPT-5.4 Pro83.3
MultimodalClaude 3.5 SonnetNot measuredMarginNo overlapGPT-5.4 Pro94.0

Decisive benchmark drivers

The largest measured benchmark gaps in this matchup, with exact reported values.

More
A · Claude 3.5 SonnetB · GPT-5.4 Pro
  1. FrontierMath v2 (Tiers 1-3)

    Math
    Source ↗
    A 2.069%B 50.000%
    Winner: GPT-5.4 ProΔ 47.9
    FrontierMath v2 (Tiers 1-3): Claude 3.5 Sonnet scored 2.069%; GPT-5.4 Pro scored 50.000%. GPT-5.4 Pro wins this benchmark.
  2. FrontierMath v2 (Tier 4)

    Math
    Source ↗
    A 0.000%B 37.500%
    Winner: GPT-5.4 ProΔ 37.5
    FrontierMath v2 (Tier 4): Claude 3.5 Sonnet scored 0.000%; GPT-5.4 Pro scored 37.500%. GPT-5.4 Pro wins this benchmark.

Operational comparison

Runtime and commercial metrics are compared only when both models have a complete sourced value.

MetricClaude 3.5 SonnetGPT-5.4 ProComparison
Input / output priceUSD per 1M tokensClaude 3.5 Sonnet$3 input / $15 outputGPT-5.4 Pro$30 input / $180 outputClaude 3.5 Sonnet has the lower combined listed price.
Generation speedtokens per secondClaude 3.5 SonnetNot availableGPT-5.4 Pro74 tok/sA complete speed comparison is not available.
First-answer latencyseconds to first tokenClaude 3.5 SonnetNot availableGPT-5.4 Pro151.79 sA complete latency comparison is not available.
Context windowmaximum listed tokensClaude 3.5 Sonnet200KGPT-5.4 Pro1.05MGPT-5.4 Pro lists the larger context window.

Benchmark Deep Dive

Agentic
BenchmarkClaude 3.5 SonnetGPT-5.4 ProResult
BrowseCompSource 89.3%Not comparable
Coding
BenchmarkClaude 3.5 SonnetGPT-5.4 ProResult
SWE-bench VerifiedSource 49%Not comparable
Reasoning
BenchmarkClaude 3.5 SonnetGPT-5.4 ProResult
ARC-AGI-2Source 83.3%Not comparable
CritPtSource 30.0%Not comparable
KnowledgeClaude 3.5 Sonnet wins
BenchmarkClaude 3.5 SonnetGPT-5.4 ProResult
GPQASource 59.4%Not comparable
HLESource 58.7%Not comparable
FrontierScienceSource 36.7%Not comparable
FrontierScience ResearchSource 36.7%Not comparable
HLE w/o toolsSource 42.7%Not comparable
MathGPT-5.4 Pro wins
BenchmarkClaude 3.5 SonnetGPT-5.4 ProResult
FrontierMath v2 (Tiers 1-3)Source 2.069%50.000%GPT-5.4 Pro leads
FrontierMath v2 (Tier 4)Source 0.000%37.500%GPT-5.4 Pro leads
IPhO 2025 (Theory)Source 93.5%Not comparable
FrontierMath (legacy)Source 50%Not comparable
Multimodal
BenchmarkClaude 3.5 SonnetGPT-5.4 ProResult
MMMU-ProSource 94%Not comparable
Frequently Asked Questions (3)

Which is better, Claude 3.5 Sonnet or GPT-5.4 Pro?

GPT-5.4 Pro is ahead on BenchLM's BenchAlign leaderboard, 60.89 to 47.74. The biggest single separator in this matchup is FrontierMath v2 (Tiers 1-3), where the scores are 2.069% and 50.000%.

Which is better for knowledge tasks, Claude 3.5 Sonnet or GPT-5.4 Pro?

Claude 3.5 Sonnet has the edge for knowledge tasks in this comparison, averaging 59.4 versus 58.7. GPT-5.4 Pro stays close enough that the answer can still flip depending on your workload.

Which is better for math, Claude 3.5 Sonnet or GPT-5.4 Pro?

GPT-5.4 Pro has the edge for math in this comparison, averaging 46.9 versus 1.6. Inside this category, FrontierMath v2 (Tiers 1-3) is the benchmark that creates the most daylight between them.

Related Comparisons

Last updated: July 23, 2026

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.