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.

SpecificationV4 ProDeepSeekClaude Opus 5Anthropic
ReleasedApr 24, 2026Jul 24, 2026
Context window1.05M1M
Max output384K128K
Input / 1M$0.435$5
Output / 1M$0.87$25
Cost per task$0.09$2.25
Arena Elo1,3131,393
Serving providers185
Parameters1598.8BUndisclosed
LicencemitProprietary
Capabilities
  • Reasoning
  • Tool use
  • 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

  • V4 Pro
  • Claude Opus 5
  • Intelligence Index

    V4 Pro
    45.3
    Claude Opus 5
    63.1
  • Coding Index

    V4 Pro
    59.4
    Claude Opus 5
    78.0
  • Agentic Index

    V4 Pro
    37.8
    Claude Opus 5
    59.2
  • Output speed

    V4 Pro
    71 t/s
    Claude Opus 5
    87 t/s
  • Context window

    V4 Pro
    1.05M
    Claude Opus 5
    1M
  • Latency · lower is better

    V4 Pro
    380ms
    Claude Opus 5
    1.12s
  • Blended price / 1M · lower is better

    V4 Pro
    $0.544
    Claude Opus 5
    $10
  • Cost per task · lower is better

    V4 Pro
    $0.09
    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
MetricV4 ProClaude Opus 5
Intelligence Index45.363.1
Coding Index59.478.0
Agentic Index37.859.2
Output speed71 t/s87 t/s
Context window1.05M1M
Latency · lower is better380ms1.12s
Blended price / 1M · lower is better$0.544$10
Cost per task · lower is better$0.09$2.25

Evaluation scores

Percentage correct on a common 0–100% scale

  • V4 Pro
  • Claude Opus 5
  • GPQA Diamond

    V4 Pro
    88.8%
    Claude Opus 5
    93.2%
  • Humanity's Last Exam

    V4 Pro
    37.5%
    Claude Opus 5
    54.9%
  • SciCode

    V4 Pro
    50.0%
    Claude Opus 5
    55.7%
  • τ²-bench

    V4 Pro
    96.2%
    Claude Opus 5
    42.1%
  • Terminal-Bench Hard

    V4 Pro
    46.2%
    Claude Opus 5not measured
  • AA-LCR long context

    V4 Pro
    70.0%
    Claude Opus 5
    75.7%

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

View as table
EvaluationV4 ProClaude Opus 5
GPQA Diamond88.8%93.2%
Humanity's Last Exam37.5%54.9%
SciCode50.0%55.7%
τ²-bench96.2%42.1%
Terminal-Bench Hard46.2%
LiveCodeBench
AA-LCR long context70.0%75.7%
AIME 2025

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