Skip to content
llmwaves

Compare

Four models, one page

Pick the models you are actually choosing between. Every row is scaled within itself, so a bar means something next to its neighbours and nothing across rows.

SpecificationQwen3.8 MaxQwenClaude Opus 5.5AnthropicClaude Sonnet 5.5Anthropic
ReleasedAug 3, 2026Sep 22, 2026Sep 28, 2026
Context window1M1M1M
Max output131K128K128K
Input / 1M$2$4$2
Output / 1M$6$20$10
Cost per task$0.58$1.80$0.90
Arena Elo—1,4791,468
Serving providers155
ParametersUndisclosedUndisclosedUndisclosed
LicenceProprietaryProprietaryProprietary
Capabilities
  • Reasoning
  • Tool use
  • Vision
  • Video in
  • Reasoning
  • Tool use
  • Vision
  • Reasoning
  • Tool use
  • Vision

Metrics side by side

Each row is scaled to the largest value in that row — bars compare within a row, never across rows

  • Qwen3.8 Max
  • Claude Opus 5.5
  • Claude Sonnet 5.5

  • Intelligence Index

    Qwen3.8 Max
    58.1
    Claude Opus 5.5
    57.6
    Claude Sonnet 5.5
    56.0

  • Coding Index

    Qwen3.8 Max
    71.8
    Claude Opus 5.5not measured
    Claude Sonnet 5.5not measured

  • Agentic Index

    Qwen3.8 Max
    58.4
    Claude Opus 5.5not measured
    Claude Sonnet 5.5not measured

  • Output speed

    Qwen3.8 Max
    38 t/s
    Claude Opus 5.5
    174 t/s
    Claude Sonnet 5.5
    118 t/s

  • Context window

    Qwen3.8 Max
    1M
    Claude Opus 5.5
    1M
    Claude Sonnet 5.5
    1M

  • Latency · lower is better

    Qwen3.8 Max
    2.9s
    Claude Opus 5.5
    2.28s
    Claude Sonnet 5.5
    902ms

  • Blended price / 1M · lower is better

    Qwen3.8 Max
    $3
    Claude Opus 5.5
    $8
    Claude Sonnet 5.5
    $4

  • Cost per task · lower is better

    Qwen3.8 Max
    $0.58
    Claude Opus 5.5
    $1.80
    Claude Sonnet 5.5
    $0.90

Rows marked “lower is better” still draw a longer bar for a larger number — read the value, not just the length. Arena Elo is in the specification table above instead: it has no meaningful zero, so a bar would flatten the gaps.

View as table
MetricQwen3.8 MaxClaude Opus 5.5Claude Sonnet 5.5
Intelligence Index58.157.656.0
Coding Index71.8——
Agentic Index58.4——
Output speed38 t/s174 t/s118 t/s
Context window1M1M1M
Latency · lower is better2.9s2.28s902ms
Blended price / 1M · lower is better$3$8$4
Cost per task · lower is better$0.58$1.80$0.90

Evaluation scores

Percentage correct on a common 0–100% scale

  • Qwen3.8 Max
  • Claude Opus 5.5
  • Claude Sonnet 5.5

  • GPQA Diamond

    Qwen3.8 Max
    92.7%
    Claude Opus 5.5not measured
    Claude Sonnet 5.5not measured

  • Humanity's Last Exam

    Qwen3.8 Max
    43.0%
    Claude Opus 5.5
    61.4%
    Claude Sonnet 5.5
    55.0%

  • SciCode

    Qwen3.8 Max
    52.9%
    Claude Opus 5.5
    66.9%
    Claude Sonnet 5.5
    61.0%

  • τ²-bench

    Qwen3.8 Max
    51.3%
    Claude Opus 5.5not measured
    Claude Sonnet 5.5not measured

  • AA-LCR long context

    Qwen3.8 Max
    74.3%
    Claude Opus 5.5
    84.7%
    Claude Sonnet 5.5
    82.7%

A missing bar means that evaluation was not run for that model — it is not a zero.

View as table
EvaluationQwen3.8 MaxClaude Opus 5.5Claude Sonnet 5.5
GPQA Diamond92.7%——
Humanity's Last Exam43.0%61.4%55.0%
SciCode52.9%66.9%61.0%
τ²-bench51.3%——
Terminal-Bench Hard———
LiveCodeBench———
AA-LCR long context74.3%84.7%82.7%
AIME 2025———

Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).