Skip to main content
BenchLM

Model comparison

Claude Opus 4.8 vs Qwen3.5 397B

Data verified

Head-to-head evidence from 28 shared benchmark results across 6 categories. Overall scores shown here use BenchLM's provisional ranking lane.

85/100
Margin
25.0pts
← winning
60/100
3 category wins2 category wins

Verified leaderboard positions: Claude Opus 4.8 #3; Qwen3.5 397B #23

Evidence parity. Claude Opus 4.8 and Qwen3.5 397B share 28 comparable benchmark results. 5 of 8 categories are comparable. 25 results are unique to Claude Opus 4.8; 28 to Qwen3.5 397B.

Updated July 12, 2026
Shared results
28
Claude Opus 4.8 only
25
Qwen3.5 397B only
28
Comparable categories
5 / 8

Pick Claude Opus 4.8 if you want the stronger benchmark profile. Qwen3.5 397B only becomes the better choice if mathematics is the priority or you want the cheaper token bill.

Confidence note. This is a partial-evidence comparison with 28 shared benchmark results across 6 evidence categories; 5 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

Claude Opus 4.8 is clearly ahead on the provisional aggregate, 85 to 60. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

Claude Opus 4.8's sharpest advantage is in agentic, where it averages 80.3 against 56.5. The single biggest benchmark swing on the page is HLE, 57.9% to 28.7%. Qwen3.5 397B does hit back in mathematics, so the answer changes if that is the part of the workload you care about most.

Claude Opus 4.8 is also the more expensive model on tokens at $5.00 input / $25.00 output per 1M tokens, versus $0.60 input / $3.60 output per 1M tokens for Qwen3.5 397B. That is roughly 6.9x on output cost alone. Claude Opus 4.8 is the reasoning model in the pair, while Qwen3.5 397B 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. Claude Opus 4.8 gives you the larger context window at 1M, compared with 128K for Qwen3.5 397B.

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 Opus 4.8 and Qwen3.5 397B
CategoryClaude Opus 4.8ΔQwen3.5 397B
MathClaude Opus 4.853.9Margin→ 36.7Qwen3.5 397B90.6
AgenticClaude Opus 4.880.3Margin← 23.8Qwen3.5 397B56.5
CodingClaude Opus 4.876.4Margin← 16.1Qwen3.5 397B60.3
KnowledgeClaude Opus 4.862.7Margin← 5.8Qwen3.5 397B56.9
MultimodalClaude Opus 4.877.0Margin→ 2.6Qwen3.5 397B79.6
ReasoningClaude Opus 4.8Not measuredMarginNo overlapQwen3.5 397B63.2
MultilingualClaude Opus 4.8Not measuredMarginNo overlapQwen3.5 397B84.7
Inst. FollowingClaude Opus 4.8Not measuredMarginNo overlapQwen3.5 397B92.6

Decisive benchmark drivers

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

More
A · Claude Opus 4.8B · Qwen3.5 397B

  1. HLE

    Knowledge
    Source ↗
    A 57.9%B 28.7%
    Winner: Claude Opus 4.8Δ 29.2
    HLE: Claude Opus 4.8 scored 57.9%; Qwen3.5 397B scored 28.7%. Claude Opus 4.8 wins this benchmark.

  2. BrowseComp

    Agentic
    Source ↗
    A 84.3%B 62%
    Winner: Claude Opus 4.8Δ 22.3
    BrowseComp: Claude Opus 4.8 scored 84.3%; Qwen3.5 397B scored 62%. Claude Opus 4.8 wins this benchmark.

  3. Terminal-Bench 2.0

    Agentic
    Source ↗
    A 74.6%B 52.5%
    Winner: Claude Opus 4.8Δ 22.1
    Terminal-Bench 2.0: Claude Opus 4.8 scored 74.6%; Qwen3.5 397B scored 52.5%. Claude Opus 4.8 wins this benchmark.

  4. SWE-bench Pro

    Coding
    Source ↗
    A 69.2%B 50.9%
    Winner: Claude Opus 4.8Δ 18.3
    SWE-bench Pro: Claude Opus 4.8 scored 69.2%; Qwen3.5 397B scored 50.9%. Claude Opus 4.8 wins this benchmark.

  5. SWE-bench Verified

    Coding
    Source ↗
    A 88.6%B 76.2%
    Winner: Claude Opus 4.8Δ 12.4
    SWE-bench Verified: Claude Opus 4.8 scored 88.6%; Qwen3.5 397B scored 76.2%. Claude Opus 4.8 wins this benchmark.

Operational comparison

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

MetricClaude Opus 4.8Qwen3.5 397BComparison
Input / output priceUSD per 1M tokensClaude Opus 4.8$5 input / $25 outputQwen3.5 397B$0.6 input / $3.6 outputQwen3.5 397B has the lower combined listed price.
Generation speedtokens per secondClaude Opus 4.8Not availableQwen3.5 397B96 tok/sA complete speed comparison is not available.
First-answer latencyseconds to first tokenClaude Opus 4.8Not availableQwen3.5 397B2.44 sA complete latency comparison is not available.
Context windowmaximum listed tokensClaude Opus 4.81MQwen3.5 397B128KClaude Opus 4.8 lists the larger context window.

Benchmark Deep Dive

AgenticClaude Opus 4.8 wins
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
Terminal-Bench 2.0Source 74.6%52.5%Claude Opus 4.8 leads
BrowseCompSource 84.3%62%Claude Opus 4.8 leads
DeepSearchQASource 93.1%—Not comparable
OSWorld-VerifiedSource 83.4%—Not comparable
Finance Agent v2Source 53.9%—Not comparable
GDPval-AASource 1600960Claude Opus 4.8 leads
MCP AtlasSource 82.2%46.1%Claude Opus 4.8 leads
ToolathlonSource 59.9%36.3%Claude Opus 4.8 leads
Gert LabsSource 72.97%46.76%Claude Opus 4.8 leads
AA Agentic IndexSource 47.2%19.9%Claude Opus 4.8 leads
Tau2-TelecomSource 94.4%95.6%Qwen3.5 397B leads
GDPval-AASource 55.0%23.0%Claude Opus 4.8 leads
ResearchClawBenchSource 21.1%14.2%Claude Opus 4.8 leads
OSWorld 2.0Source 20.6%—Not comparable
AA BriefcaseSource 1354—Not comparable
AA AutomationBenchSource 48.5%—Not comparable
AA EnterpriseOps-GymSource 44.0%—Not comparable
AA Harvey LABSource 7.5%—Not comparable
AA Tau3 BankingSource 27.6%—Not comparable
Claw-EvalSource —56.8%Not comparable
QwenClawBenchSource —51.8%Not comparable
TAU3-BenchSource —68.4%Not comparable
VITA-BenchSource —43.7%Not comparable
DeepPlanningSource —37.6%Not comparable
MCP-TasksSource —74.2%Not comparable
WideResearchSource —74.0%Not comparable
APEX-Agents-AASource —15.3%Not comparable
CodingClaude Opus 4.8 wins
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
SWE-bench VerifiedSource 88.6%76.2%Claude Opus 4.8 leads
SWE-bench ProSource 69.2%50.9%Claude Opus 4.8 leads
SWE MultilingualSource 84.4%—Not comparable
SWE MultimodalSource 38.4%—Not comparable
Terminal-Bench 2.0Source 74.6%—Not comparable
cursorBench31Source 58.4%—Not comparable
cursorBench32Source 62.3%—Not comparable
AA Coding IndexSource 74.3%48.2%Claude Opus 4.8 leads
Terminal-Bench HardSource 58.3%40.9%Claude Opus 4.8 leads
AA-SciCodeSource 53.5%42.0%Claude Opus 4.8 leads
FrontierCodeSource 46.5%—Not comparable
AA Terminal-Bench 2.1Source 84.6%—Not comparable
LiveCodeBench v6Source —83.6%Not comparable
Reasoning
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
AA-LCRSource 67.7%65.7%Claude Opus 4.8 leads
CritPtSource 20.9%1.7%Claude Opus 4.8 leads
LongBench v2Source —63.2%Not comparable
AI-NeedleSource —68.7%Not comparable
KnowledgeClaude Opus 4.8 wins
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
GPQASource 93.6%88.4%Claude Opus 4.8 leads
GPQA-DSource 93.6%—Not comparable
HLESource 57.9%28.7%Claude Opus 4.8 leads
HLE w/o toolsSource 49.8%—Not comparable
Artificial Analysis Intelligence IndexSource 55.7%33.7%Claude Opus 4.8 leads
AA-GPQA DiamondSource 92.0%89.3%Claude Opus 4.8 leads
AA-HLESource 45.7%27.3%Claude Opus 4.8 leads
AA-Omniscience IndexSource 27.4%-29.8%Claude Opus 4.8 leads
AA-Omniscience AccuracySource 46.6%31.4%Claude Opus 4.8 leads
AA-Omniscience Hallucination RateSource 35.9%89.1%Claude Opus 4.8 leads
SuperGPQASource —70.4%Not comparable
MMLU-ProSource —87.8%Not comparable
MMLU-ReduxSource —94.9%Not comparable
C-EvalSource —93%Not comparable
MathQwen3.5 397B wins
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
USAMO 2026Source 96.7%—Not comparable
FrontierMath v2 (Tiers 1-3)Source 47.241%—Not comparable
FrontierMath v2 (Tier 4)Source 31.250%—Not comparable
AIME26Source —93.3%Not comparable
HMMT Feb 2025Source —94.8%Not comparable
HMMT Nov 2025Source —92.7%Not comparable
HMMT Feb 2026Source —87.9%Not comparable
MMAnswerBenchSource —80.9%Not comparable
Multilingual
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
INCLUDESource 87.6%—Not comparable
MMLU-ProXSource —84.7%Not comparable
NOVA-63Source —59.1%Not comparable
MultimodalQwen3.5 397B wins
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
OfficeQA ProSource 66.2%—Not comparable
ScreenSpot ProSource 87.9%65.6%Claude Opus 4.8 leads
CharXivSource 89.9%80.8%Claude Opus 4.8 leads
CharXiv w/o toolsSource 80.5%—Not comparable
Design Arena WebsiteSource 1281—Not comparable
MMMU-ProSource —79%Not comparable
MathVisionSource —88.6%Not comparable
VideoMMMUSource —84.7%Not comparable
V*Source —95.8%Not comparable
AA-MMMU-ProSource —77.3%Not comparable
Inst. Following
BenchmarkClaude Opus 4.8Qwen3.5 397BResult
AA-IFBenchSource 62.2%78.8%Qwen3.5 397B leads
IFEvalSource —92.6%Not comparable
Frequently Asked Questions (6)

Which is better, Claude Opus 4.8 or Qwen3.5 397B?

Claude Opus 4.8 is ahead on BenchLM's provisional leaderboard, 85 to 60. The biggest single separator in this matchup is HLE, where the scores are 57.9% and 28.7%.

Which is better for knowledge tasks, Claude Opus 4.8 or Qwen3.5 397B?

Claude Opus 4.8 has the edge for knowledge tasks in this comparison, averaging 62.7 versus 56.9. Inside this category, AA-Omniscience Index is the benchmark that creates the most daylight between them.

Which is better for coding, Claude Opus 4.8 or Qwen3.5 397B?

Claude Opus 4.8 has the edge for coding in this comparison, averaging 76.4 versus 60.3. Inside this category, AA Coding Index is the benchmark that creates the most daylight between them.

Which is better for math, Claude Opus 4.8 or Qwen3.5 397B?

Qwen3.5 397B has the edge for math in this comparison, averaging 90.6 versus 53.9. Claude Opus 4.8 stays close enough that the answer can still flip depending on your workload.

Which is better for agentic tasks, Claude Opus 4.8 or Qwen3.5 397B?

Claude Opus 4.8 has the edge for agentic tasks in this comparison, averaging 80.3 versus 56.5. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.

Which is better for multimodal and grounded tasks, Claude Opus 4.8 or Qwen3.5 397B?

Qwen3.5 397B has the edge for multimodal and grounded tasks in this comparison, averaging 79.6 versus 77. Inside this category, ScreenSpot Pro 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 12, 2026

The AI models change fast. We track them for you.

A weekly brief for engineers and researchers covering new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.