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.

SpecificationPhi 4MicrosoftClaude Opus 5Anthropic
ReleasedJan 10, 2025Jul 24, 2026
Context window16K1M
Max output16K128K
Input / 1M$0.07$5
Output / 1M$0.14$25
Cost per task$0.007$2.25
Arena Elo—1,393
Serving providers15
Parameters14.7BUndisclosed
LicencemitProprietary
Capabilities
  • 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

  • Phi 4
  • Claude Opus 5

  • Intelligence Index

    Phi 4
    4.6
    Claude Opus 5
    63.1

  • Coding Index

    Phi 4not measured
    Claude Opus 5
    78.0

  • Agentic Index

    Phi 4not measured
    Claude Opus 5
    59.2

  • Output speed

    Phi 4
    76 t/s
    Claude Opus 5
    87 t/s

  • Context window

    Phi 4
    16K
    Claude Opus 5
    1M

  • Latency · lower is better

    Phi 4
    304ms
    Claude Opus 5
    1.12s

  • Blended price / 1M · lower is better

    Phi 4
    $0.088
    Claude Opus 5
    $10

  • Cost per task · lower is better

    Phi 4
    $0.007
    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
MetricPhi 4Claude Opus 5
Intelligence Index4.663.1
Coding Index—78.0
Agentic Index—59.2
Output speed76 t/s87 t/s
Context window16K1M
Latency · lower is better304ms1.12s
Blended price / 1M · lower is better$0.088$10
Cost per task · lower is better$0.007$2.25

Evaluation scores

Percentage correct on a common 0–100% scale

  • Phi 4
  • Claude Opus 5

  • GPQA Diamond

    Phi 4
    57.5%
    Claude Opus 5
    93.2%

  • Humanity's Last Exam

    Phi 4
    3.8%
    Claude Opus 5
    54.9%

  • SciCode

    Phi 4
    26.0%
    Claude Opus 5
    55.7%

  • τ²-bench

    Phi 4
    0.0%
    Claude Opus 5
    42.1%

  • Terminal-Bench Hard

    Phi 4
    3.8%
    Claude Opus 5not measured

  • LiveCodeBench

    Phi 4
    23.1%
    Claude Opus 5not measured

  • AA-LCR long context

    Phi 4
    0.0%
    Claude Opus 5
    75.7%

  • AIME 2025

    Phi 4
    18.0%
    Claude Opus 5not measured

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

View as table
EvaluationPhi 4Claude Opus 5
GPQA Diamond57.5%93.2%
Humanity's Last Exam3.8%54.9%
SciCode26.0%55.7%
τ²-bench0.0%42.1%
Terminal-Bench Hard3.8%—
LiveCodeBench23.1%—
AA-LCR long context0.0%75.7%
AIME 202518.0%—

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