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 VL 30B A3B InstructQwenClaude Opus 5Anthropic
ReleasedOct 6, 2025Jul 24, 2026
Context window262K1M
Max output16K128K
Input / 1M$0.15$5
Output / 1M$0.6$25
Cost per task$0.02$2.25
Arena Elo—1,393
Serving providers45
Parameters31.1BUndisclosed
Licenceapache-2.0Proprietary
Capabilities
  • Tool use
  • Vision
  • Open weights
  • 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 VL 30B A3B Instruct
  • Claude Opus 5

  • Intelligence Index

    Qwen3 VL 30B A3B Instruct
    9.9
    Claude Opus 5
    63.1

  • Coding Index

    Qwen3 VL 30B A3B Instructnot measured
    Claude Opus 5
    78.0

  • Agentic Index

    Qwen3 VL 30B A3B Instructnot measured
    Claude Opus 5
    59.2

  • Output speed

    Qwen3 VL 30B A3B Instruct
    16 t/s
    Claude Opus 5
    87 t/s

  • Context window

    Qwen3 VL 30B A3B Instruct
    262K
    Claude Opus 5
    1M

  • Latency · lower is better

    Qwen3 VL 30B A3B Instruct
    417ms
    Claude Opus 5
    1.12s

  • Blended price / 1M · lower is better

    Qwen3 VL 30B A3B Instruct
    $0.262
    Claude Opus 5
    $10

  • Cost per task · lower is better

    Qwen3 VL 30B A3B Instruct
    $0.02
    Claude Opus 5
    $2.25

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 VL 30B A3B InstructClaude Opus 5
Intelligence Index9.963.1
Coding Index—78.0
Agentic Index—59.2
Output speed16 t/s87 t/s
Context window262K1M
Latency · lower is better417ms1.12s
Blended price / 1M · lower is better$0.262$10
Cost per task · lower is better$0.02$2.25

Evaluation scores

Percentage correct on a common 0–100% scale

  • Qwen3 VL 30B A3B Instruct
  • Claude Opus 5

  • GPQA Diamond

    Qwen3 VL 30B A3B Instruct
    69.5%
    Claude Opus 5
    93.2%

  • Humanity's Last Exam

    Qwen3 VL 30B A3B Instruct
    6.3%
    Claude Opus 5
    54.9%

  • SciCode

    Qwen3 VL 30B A3B Instruct
    30.8%
    Claude Opus 5
    55.7%

  • τ²-bench

    Qwen3 VL 30B A3B Instruct
    19.0%
    Claude Opus 5
    42.1%

  • Terminal-Bench Hard

    Qwen3 VL 30B A3B Instruct
    6.1%
    Claude Opus 5not measured

  • LiveCodeBench

    Qwen3 VL 30B A3B Instruct
    47.6%
    Claude Opus 5not measured

  • AA-LCR long context

    Qwen3 VL 30B A3B Instruct
    27.3%
    Claude Opus 5
    75.7%

  • AIME 2025

    Qwen3 VL 30B A3B Instruct
    72.3%
    Claude Opus 5not measured

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

View as table
EvaluationQwen3 VL 30B A3B InstructClaude Opus 5
GPQA Diamond69.5%93.2%
Humanity's Last Exam6.3%54.9%
SciCode30.8%55.7%
τ²-bench19.0%42.1%
Terminal-Bench Hard6.1%—
LiveCodeBench47.6%—
AA-LCR long context27.3%75.7%
AIME 202572.3%—

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