Skip to main content
BenchLM
Advertise

Model comparison

Claude Sonnet 4.5 vs Claude Sonnet 4.6

Data verified

Head-to-head evidence from 9 shared benchmark results across 5 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.

53.61/100
Margin
11.5pts
winning →
65.07/100
2 category wins2 category wins

Public leaderboard positions: Claude Sonnet 4.5 #96 (Estimated); Claude Sonnet 4.6 #32 (Supported). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. Claude Sonnet 4.5 and Claude Sonnet 4.6 share 9 comparable benchmark results. 4 of 8 categories are comparable. 3 results are unique to Claude Sonnet 4.5; 24 to Claude Sonnet 4.6.

Updated July 20, 2026
Shared results
9
Claude Sonnet 4.5 only
3
Claude Sonnet 4.6 only
24
Comparable categories
4 / 8

Pick Claude Sonnet 4.6 if you want the stronger benchmark profile. Claude Sonnet 4.5 only becomes the better choice if knowledge is the priority.

Confidence note. This is a partial-evidence comparison with 9 shared benchmark results across 5 evidence categories; 4 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

Claude Sonnet 4.6 is clearly ahead on the BenchAlign aggregate, 65.07 to 53.61. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

Claude Sonnet 4.6's sharpest advantage is in mathematics, where it averages 26.4 against 11.2. The single biggest benchmark swing on the page is FrontierMath v2 (Tiers 1-3), 13.495% to 32.400%. Claude Sonnet 4.5 does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.

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 Sonnet 4.5 and Claude Sonnet 4.6
CategoryClaude Sonnet 4.5ΔClaude Sonnet 4.6
KnowledgeClaude Sonnet 4.583.4Margin← 17.4Claude Sonnet 4.666.0
MathClaude Sonnet 4.511.2Margin→ 15.2Claude Sonnet 4.626.4
AgenticClaude Sonnet 4.555.4Margin→ 9.8Claude Sonnet 4.665.2
CodingClaude Sonnet 4.577.2Margin← 8.1Claude Sonnet 4.669.1
ReasoningClaude Sonnet 4.513.6MarginNo overlapClaude Sonnet 4.6Not measured
MultimodalClaude Sonnet 4.5Not measuredMarginNo overlapClaude Sonnet 4.677.4

Decisive benchmark drivers

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

More
A · Claude Sonnet 4.5B · Claude Sonnet 4.6

  1. FrontierMath v2 (Tiers 1-3)

    Math
    Source ↗
    A 13.495%B 32.400%
    Winner: Claude Sonnet 4.6Δ 18.9
    FrontierMath v2 (Tiers 1-3): Claude Sonnet 4.5 scored 13.495%; Claude Sonnet 4.6 scored 32.400%. Claude Sonnet 4.6 wins this benchmark.

  2. OSWorld-Verified

    Agentic
    Source ↗
    A 61.4%B 72.1%
    Winner: Claude Sonnet 4.6Δ 10.7
    OSWorld-Verified: Claude Sonnet 4.5 scored 61.4%; Claude Sonnet 4.6 scored 72.1%. Claude Sonnet 4.6 wins this benchmark.

  3. Terminal-Bench 2.0

    Agentic
    Source ↗
    A 50%B 59.1%
    Winner: Claude Sonnet 4.6Δ 9.1
    Terminal-Bench 2.0: Claude Sonnet 4.5 scored 50%; Claude Sonnet 4.6 scored 59.1%. Claude Sonnet 4.6 wins this benchmark.

  4. GPQA

    Knowledge
    Source ↗
    A 83.4%B 89.9%
    Winner: Claude Sonnet 4.6Δ 6.5
    GPQA: Claude Sonnet 4.5 scored 83.4%; Claude Sonnet 4.6 scored 89.9%. Claude Sonnet 4.6 wins this benchmark.

  5. FrontierMath v2 (Tier 4)

    Math
    Source ↗
    A 4.167%B 8.300%
    Winner: Claude Sonnet 4.6Δ 4.1
    FrontierMath v2 (Tier 4): Claude Sonnet 4.5 scored 4.167%; Claude Sonnet 4.6 scored 8.300%. Claude Sonnet 4.6 wins this benchmark.

Operational comparison

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

MetricClaude Sonnet 4.5Claude Sonnet 4.6Comparison
Input / output priceUSD per 1M tokensClaude Sonnet 4.5$3 input / $15 outputClaude Sonnet 4.6$3 input / $15 outputListed prices are equal.
Generation speedtokens per secondClaude Sonnet 4.5Not availableClaude Sonnet 4.644 tok/sA complete speed comparison is not available.
First-answer latencyseconds to first tokenClaude Sonnet 4.5Not availableClaude Sonnet 4.61.48 sA complete latency comparison is not available.
Context windowmaximum listed tokensClaude Sonnet 4.5200KClaude Sonnet 4.6200KListed context windows are equal.

Benchmark Deep Dive

AgenticClaude Sonnet 4.6 wins
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
Terminal-Bench 2.0Source 50%59.1%Claude Sonnet 4.6 leads
OSWorld-VerifiedSource 61.4%72.1%Claude Sonnet 4.6 leads
VITA-BenchSource 17.0%—Not comparable
Gert LabsSource 48.51%62.92%Claude Sonnet 4.6 leads
JobBenchSource 27.7%36.9%Claude Sonnet 4.6 leads
Claw-EvalSource —67.8%Not comparable
CyberGymSource —65.2%Not comparable
τ²-bench resultsSource —79.5%Not comparable
OSWorld 2.0Source —8.3%Not comparable
CodingClaude Sonnet 4.5 wins
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
SWE-bench VerifiedSource 77.2%79.6%Claude Sonnet 4.6 leads
SWE-RebenchSource —60.7%Not comparable
React Native EvalsSource —80.6%Not comparable
Vibe Code BenchSource —51.48%Not comparable
cursorBench31Source —48.8%Not comparable
AA-SciCodeSource —46.9%Not comparable
FrontierCode 1.1 MainSource —24.3%Not comparable
Reasoning
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
ARC-AGI-2Source 13.6%—Not comparable
AA-LCRSource —57.7%Not comparable
CritPtSource —0.9%Not comparable
KnowledgeClaude Sonnet 4.5 wins
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
GPQASource 83.4%89.9%Claude Sonnet 4.6 leads
SuperGPQASource —95%Not comparable
MMLU-ProSource —79.2%Not comparable
HLESource —49%Not comparable
Artificial Analysis Intelligence IndexSource —35.9%Not comparable
AA-GPQA DiamondSource —79.9%Not comparable
AA-HLESource —13.2%Not comparable
AA-Omniscience IndexSource —-2.9%Not comparable
AA-Omniscience AccuracySource —38.0%Not comparable
AA-Omniscience Hallucination RateSource —65.9%Not comparable
MathClaude Sonnet 4.6 wins
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
AIME 2025Source 87%—Not comparable
FrontierMath v2 (Tiers 1-3)Source 13.495%32.400%Claude Sonnet 4.6 leads
FrontierMath v2 (Tier 4)Source 4.167%8.300%Claude Sonnet 4.6 leads
Multimodal
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
Design Arena WebsiteSource 12221317Claude Sonnet 4.6 leads
CharXivSource —77.4%Not comparable
AA-MMMU-ProSource —70.6%Not comparable
Inst. Following
BenchmarkClaude Sonnet 4.5Claude Sonnet 4.6Result
AA-IFBenchSource —41.2%Not comparable
Frequently Asked Questions (5)

Which is better, Claude Sonnet 4.5 or Claude Sonnet 4.6?

Claude Sonnet 4.6 is ahead on BenchLM's BenchAlign leaderboard, 65.07 to 53.61. The biggest single separator in this matchup is FrontierMath v2 (Tiers 1-3), where the scores are 13.495% and 32.400%.

Which is better for knowledge tasks, Claude Sonnet 4.5 or Claude Sonnet 4.6?

Claude Sonnet 4.5 has the edge for knowledge tasks in this comparison, averaging 83.4 versus 66. Inside this category, GPQA is the benchmark that creates the most daylight between them.

Which is better for coding, Claude Sonnet 4.5 or Claude Sonnet 4.6?

Claude Sonnet 4.5 has the edge for coding in this comparison, averaging 77.2 versus 69.1. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.

Which is better for math, Claude Sonnet 4.5 or Claude Sonnet 4.6?

Claude Sonnet 4.6 has the edge for math in this comparison, averaging 26.4 versus 11.2. Inside this category, FrontierMath v2 (Tiers 1-3) is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, Claude Sonnet 4.5 or Claude Sonnet 4.6?

Claude Sonnet 4.6 has the edge for agentic tasks in this comparison, averaging 65.2 versus 55.4. Inside this category, Gert Labs is the benchmark that creates the most daylight between them.

Related Comparisons

Explore More

Anthropic Compare Pricing Methodology Find Your Best LLM Overall Rankings
Last updated: July 20, 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.