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.

SpecificationV3.2DeepSeekClaude Opus 5Anthropic
ReleasedDec 1, 2025Jul 24, 2026
Context window164K1M
Max output66K128K
Input / 1M$0.269$5
Output / 1M$0.4$25
Cost per task$0.05$2.25
Arena Elo1,393
Serving providers145
Parameters685.4BUndisclosed
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

  • V3.2
  • Claude Opus 5
  • Intelligence Index

    V3.2
    25.1
    Claude Opus 5
    63.1
  • Coding Index

    V3.2not measured
    Claude Opus 5
    78.0
  • Agentic Index

    V3.2not measured
    Claude Opus 5
    59.2
  • Output speed

    V3.2
    100 t/s
    Claude Opus 5
    87 t/s
  • Context window

    V3.2
    164K
    Claude Opus 5
    1M
  • Latency · lower is better

    V3.2
    676ms
    Claude Opus 5
    1.12s
  • Blended price / 1M · lower is better

    V3.2
    $0.302
    Claude Opus 5
    $10
  • Cost per task · lower is better

    V3.2
    $0.05
    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
MetricV3.2Claude Opus 5
Intelligence Index25.163.1
Coding Index78.0
Agentic Index59.2
Output speed100 t/s87 t/s
Context window164K1M
Latency · lower is better676ms1.12s
Blended price / 1M · lower is better$0.302$10
Cost per task · lower is better$0.05$2.25

Evaluation scores

Percentage correct on a common 0–100% scale

  • V3.2
  • Claude Opus 5
  • GPQA Diamond

    V3.2
    75.1%
    Claude Opus 5
    93.2%
  • Humanity's Last Exam

    V3.2
    11.2%
    Claude Opus 5
    54.9%
  • SciCode

    V3.2
    38.7%
    Claude Opus 5
    55.7%
  • τ²-bench

    V3.2
    78.9%
    Claude Opus 5
    42.1%
  • Terminal-Bench Hard

    V3.2
    32.6%
    Claude Opus 5not measured
  • LiveCodeBench

    V3.2
    59.3%
    Claude Opus 5not measured
  • AA-LCR long context

    V3.2
    42.7%
    Claude Opus 5
    75.7%
  • AIME 2025

    V3.2
    59.0%
    Claude Opus 5not measured

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

View as table
EvaluationV3.2Claude Opus 5
GPQA Diamond75.1%93.2%
Humanity's Last Exam11.2%54.9%
SciCode38.7%55.7%
τ²-bench78.9%42.1%
Terminal-Bench Hard32.6%
LiveCodeBench59.3%
AA-LCR long context42.7%75.7%
AIME 202559.0%

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