LogoLMSpeed
Back to models

Model compare

Claude Opus 4.8 vs GPT-5.4

The readout for Claude Opus 4.8 and GPT-5.4, before the detailed comparison sheet.

Model A

Claude Opus 4.8

Anthropic

Leading
vs

Model B

GPT-5.4

OpenAI

Contender

Key Takeaways

Weighted outcome: Claude Opus 4.8. Benchmark capability categories carry 80%, while price, API performance, and availability carry 20%.

Decision read

Claude Opus 4.8

Claude Opus 4.8 has the higher weighted result; Model A / B score 76.6 to 23.4.

Evidence depth

160 data points

Includes 48 benchmark rows, 0 audit samples, and 10 provider examples.

Selection signal

Start with Claude Opus 4.8

The charts below split 58 high-signal samples across speed, scores, and audit health.

Change comparison

Switch either side of this report to compare another model with the same LMSpeed data pipeline.

Select a different model to open a new comparison URL.

Comparison sheet

This report only uses LMSpeed data for Claude Opus 4.8 and GPT-5.4: pricing, speed aggregates, third-party benchmark scores, and shared provider samples.

Model compare
Claude Opus 4.8
GPT-5.4
Overall leaderLeadingContender
Weighted overall score76.6 pts23.4 pts
Benchmark category leads6 categories1 categories
Operational advantagesAverage speed, First-token latencyCheapest input price, Free providers, Provider coverage
Context window1M tokens1.1M tokens
Max output128K tokens128K tokens
Modalities

Input

TextImageFile

Output

Text

Input

TextImageFile

Output

Text
Features
Text inputImage inputFile inputText outputTool callingStructured outputsJSON modeReasoning
Text inputImage inputFile inputText outputTool callingStructured outputsJSON modeReasoning

The overall result weights benchmark capability categories at 80% and price, API speed/latency, and availability at 20%. Recent test volume does not affect the winner, and missing benchmark categories are excluded.

Model metadata

Model compare
Claude Opus 4.8
GPT-5.4
DeveloperAnthropicOpenAI
ReleasedMay 2026Mar 2026
ParametersNo dataNo data
TokenizerClaudeGPT
Knowledge cutoffNo dataNo data
OpenRouter IDanthropic/claude-opus-4.8openai/gpt-5.4
ReferencesNo dataNo data

When to choose each model

This report only uses LMSpeed data for Claude Opus 4.8 and GPT-5.4: pricing, speed aggregates, third-party benchmark scores, and shared provider samples.

Claude Opus 4.8

Claude Opus 4.8 is stronger in benchmark categories (Agents, Coding, Reasoning, Knowledge, Math) and operational dimensions (Average speed, First-token latency).

GPT-5.4

GPT-5.4 is stronger in benchmark categories (Instruction following) and operational dimensions (Cheapest input price, Free providers, Provider coverage).

Benchmark score comparison

Third-party benchmark profile synced into LMSpeed; only metrics available for both models are shown.

Category performance

Compare benchmark category scores on a 0-100 scale. Select a category to inspect the gap.

Model A coverage
8 / 8
Model B coverage
7 / 8
Shared
7 shared categories

Avg. score

Claude Opus 4.8

58.1

Avg. score

GPT-5.4

56.6

Agents

Claude Opus 4.8 leads by 5.5

Claude Opus 4.861.8
GPT-5.456.3

Coding

Claude Opus 4.8 leads by 1.8

Claude Opus 4.864
GPT-5.462.2

Reasoning

Claude Opus 4.8 leads by 0.2

Claude Opus 4.855.9
GPT-5.455.7

Knowledge

Claude Opus 4.8 leads by 3.6

Claude Opus 4.862.1
GPT-5.458.5

Math

Claude Opus 4.8 leads by 0.4

Claude Opus 4.857.2
GPT-5.456.8

Multilingual

Claude Opus 4.8

Claude Opus 4.855
GPT-5.4-

Multimodal

Claude Opus 4.8 leads by 6.5

Claude Opus 4.859
GPT-5.452.5

Instruction following

GPT-5.4 leads by 4.0

Claude Opus 4.850
GPT-5.454

Professional benchmark details

Metric-level scores with benchmark source, rank depth, confidence, error, and evaluation date where available.

Aggregatereported

BenchLM overall score

SourceClaude Opus 4.8

Claude Opus 4.8

winner

73.0

Rank #10/111 · confidence 4 · eval date 2026-05-28

+10.0

GPT-5.4

63.0

Rank #40/111 · confidence 4 · eval date 2026-03-05

Pricingverified

Input price

SourceClaude Opus 4.8 (Adaptive Reasoning, Max Effort)

Claude Opus 4.8

$5.00/M

Rank #168/186 · confidence 4

+$2.50/M

GPT-5.4

winner

$2.50/M

Rank #154/186 · confidence 4

Pricingverified

Blended price

SourceClaude Opus 4.8 (Adaptive Reasoning, Max Effort)

Claude Opus 4.8

$10.00/M

Rank #169/186 · confidence 4

+$4.38/M

GPT-5.4

winner

$5.63/M

Rank #158/186 · confidence 4

Pricingverified

Output price

SourceClaude Opus 4.8 (Adaptive Reasoning, Max Effort)

Claude Opus 4.8

$25.00/M

Rank #169/186 · confidence 4

+$10.00/M

GPT-5.4

winner

$15.00/M

Rank #159/186 · confidence 4

Reasoningreported

ARC-AGI-3

SourceClaude Opus 4.8

Claude Opus 4.8

winner

1.5

Rank #4/12 · confidence 4 · eval date 2026-05-28

+1.3

GPT-5.4

0.2

Rank #9/12 · confidence 4 · eval date 2026-03-05

Reasoningverified

HLE

SourceClaude Opus 4.8 (Adaptive Reasoning, Max Effort)

Claude Opus 4.8

winner

48.7%

Rank #7/216 · confidence 4

+37.4%

GPT-5.4

11.3%

Rank #116/216 · confidence 4

Reasoningreported

ARC-AGI-2

SourceClaude Opus 4.8

Claude Opus 4.8

72.1

Rank #12/19 · confidence 4 · eval date 2026-05-28

+1.9

GPT-5.4

winner

74.0

Rank #10/19 · confidence 4 · eval date 2026-03-05

Reasoningreported

CritPt

SourceClaude Opus 4.8

Claude Opus 4.8

20.9

Rank #13/88 · confidence 4 · eval date 2026-05-28

+2.5

GPT-5.4

winner

23.4

Rank #11/88 · confidence 4 · eval date 2026-03-05

Reasoningreported

AA-LCR

SourceClaude Opus 4.8

Claude Opus 4.8

77.7

Rank #37/106 · confidence 4 · eval date 2026-05-28

+4.3

GPT-5.4

winner

82.0

Rank #14/106 · confidence 4 · eval date 2026-03-05

Reasoningverified

GPQA

SourceClaude Opus 4.8 (Adaptive Reasoning, Max Effort)

Claude Opus 4.8

winner

92.0%

Rank #19/218 · confidence 4

+17.2%

GPT-5.4

74.8%

Rank #122/218 · confidence 4

Reasoningreported

BenchLM Reasoning score

SourceClaude Opus 4.8

Claude Opus 4.8

55.5

Rank #53/67 · confidence 4 · eval date 2026-05-28

+1.7

GPT-5.4

winner

57.2

Rank #51/67 · confidence 4 · eval date 2026-03-05

Codingreported

EEBench

SourceClaude Opus 4.8

Claude Opus 4.8

winner

51.4

Rank #3/25 · confidence 3 · eval date 2026-05-28

+12.9

GPT-5.4

38.5

Rank #9/25 · confidence 3 · eval date 2026-03-05

Codingreported

SWE-bench Pro

SourceClaude Opus 4.8

Claude Opus 4.8

winner

69.2

Rank #4/50 · confidence 4 · eval date 2026-05-28

+11.5

GPT-5.4

57.7

Rank #21/50 · confidence 4 · eval date 2026-03-05

Codingreported

AA-SciCode

SourceClaude Opus 4.8

Claude Opus 4.8

54.4

Rank #21/113 · confidence 4 · eval date 2026-05-28

+2.2

GPT-5.4

winner

56.6

Rank #10/113 · confidence 4 · eval date 2026-03-05

Codingreported

BenchLM Coding score

SourceClaude Opus 4.8

Claude Opus 4.8

winner

73.1

Rank #10/87 · confidence 4 · eval date 2026-05-28

+30.3

GPT-5.4

42.8

Rank #69/87 · confidence 4 · eval date 2026-03-05

Codingreported

AA Coding Index

SourceClaude Opus 4.8

Claude Opus 4.8

winner

74.3

Rank #13/75 · confidence 4 · eval date 2026-05-28

+3.2

GPT-5.4

71.0

Rank #21/75 · confidence 4 · eval date 2026-03-05

Mathreported

FrontierMath v2 (Tiers 1-3)

SourceClaude Opus 4.8

Claude Opus 4.8

47.2

Rank #8/47 · confidence 4 · eval date 2026-05-28

+0.4

GPT-5.4

winner

47.6

Rank #7/47 · confidence 4 · eval date 2026-03-05

Mathreported

FrontierMath v2 (Tier 4)

SourceClaude Opus 4.8

Claude Opus 4.8

winner

31.3

Rank #8/36 · confidence 4 · eval date 2026-05-28

+4.1

GPT-5.4

27.1

Rank #9/36 · confidence 4 · eval date 2026-03-05

Mathreported

BenchLM Math score

SourceClaude Opus 4.8

Claude Opus 4.8

winner

65.4

Rank #23/63 · confidence 4 · eval date 2026-05-28

+0.9

GPT-5.4

64.5

Rank #24/63 · confidence 4 · eval date 2026-03-05

Knowledgereported

HLE w/o tools

SourceClaude Opus 4.8

Claude Opus 4.8

winner

49.8

Rank #4/22 · confidence 4 · eval date 2026-05-28

+10.0

GPT-5.4

39.8

Rank #15/22 · confidence 4 · eval date 2026-03-05

Knowledgereported

GPQA-D

SourceClaude Opus 4.8

Claude Opus 4.8

winner

93.6

Rank #5/32 · confidence 4 · eval date 2026-05-28

+0.8

GPT-5.4

92.8

Rank #9/32 · confidence 4 · eval date 2026-03-05

Knowledgereported

AA-HLE

SourceClaude Opus 4.8

Claude Opus 4.8

winner

48.7

Rank #6/114 · confidence 4 · eval date 2026-05-28

+5.0

GPT-5.4

43.7

Rank #15/114 · confidence 4 · eval date 2026-03-05

Knowledgereported

BenchLM Knowledge score

SourceClaude Opus 4.8

Claude Opus 4.8

winner

86.3

Rank #6/83 · confidence 4 · eval date 2026-05-28

+12.0

GPT-5.4

74.3

Rank #30/83 · confidence 4 · eval date 2026-03-05

Knowledgereported

AA-Omniscience Index

SourceClaude Opus 4.8

Claude Opus 4.8

winner

28.8

Rank #8/45 · confidence 4 · eval date 2026-05-28

+23.0

GPT-5.4

5.8

Rank #28/45 · confidence 4 · eval date 2026-03-05

Knowledgereported

AA-Omniscience Hallucination Rate

SourceClaude Opus 4.8

Claude Opus 4.8

39.3

Rank #88/106 · confidence 4 · eval date 2026-05-28

+52.4

GPT-5.4

winner

91.7

Rank #13/106 · confidence 4 · eval date 2026-03-05

Knowledgereported

AA-Omniscience Accuracy

SourceClaude Opus 4.8

Claude Opus 4.8

48.8

Rank #19/106 · confidence 4 · eval date 2026-05-28

+2.0

GPT-5.4

winner

50.8

Rank #15/106 · confidence 4 · eval date 2026-03-05

Knowledgereported

Artificial Analysis Intelligence Index

SourceClaude Opus 4.8

Claude Opus 4.8

winner

42.0

Rank #18/117 · confidence 4 · eval date 2026-05-28

+3.0

GPT-5.4

39.0

Rank #28/117 · confidence 4 · eval date 2026-03-05

Knowledgereported

AA-GPQA Diamond

SourceClaude Opus 4.8

Claude Opus 4.8

92.0

Rank #21/113 · confidence 4 · eval date 2026-05-28

0.0

GPT-5.4

92.0

Rank #21/113 · confidence 4 · eval date 2026-03-05

Multimodalreported

OfficeQA Pro

SourceClaude Opus 4.8

Claude Opus 4.8

winner

66.2

Rank #2/10 · confidence 4 · eval date 2026-05-28

+13.0

GPT-5.4

53.2

Rank #8/10 · confidence 4 · eval date 2026-03-05

Multimodalreported

ScreenSpot Pro

SourceClaude Opus 4.8

Claude Opus 4.8

winner

87.9

Rank #2/12 · confidence 4 · eval date 2026-05-28

+2.5

GPT-5.4

85.4

Rank #3/12 · confidence 4 · eval date 2026-03-05

Multimodalreported

BenchLM Multimodal Grounded score

SourceClaude Opus 4.8

Claude Opus 4.8

winner

87.8

Rank #4/56 · confidence 4 · eval date 2026-05-28

+18.5

GPT-5.4

69.3

Rank #28/56 · confidence 4 · eval date 2026-03-05

Multimodalreported

CharXiv

SourceClaude Opus 4.8

Claude Opus 4.8

winner

89.9

Rank #5/30 · confidence 4 · eval date 2026-05-28

+7.1

GPT-5.4

82.8

Rank #12/30 · confidence 4 · eval date 2026-03-05

Multimodalreported

Design Arena Website

SourceClaude Opus 4.8

Claude Opus 4.8

winner

1269.0

Rank #31/78 · confidence 4 · eval date 2026-05-28

+35.0

GPT-5.4

1234.0

Rank #42/78 · confidence 4 · eval date 2026-03-05

Instruction followingreported

BenchLM Instruction Following score

SourceClaude Opus 4.8

Claude Opus 4.8

75.4

Rank #43/52 · confidence 4 · eval date 2026-05-28

+15.2

GPT-5.4

winner

90.6

Rank #16/52 · confidence 4 · eval date 2026-03-05

Instruction followingreported

AA-IFBench

SourceClaude Opus 4.8

Claude Opus 4.8

62.2

Rank #48/84 · confidence 4 · eval date 2026-05-28

+11.7

GPT-5.4

winner

73.9

Rank #21/84 · confidence 4 · eval date 2026-03-05

Agentsreported

Gert Labs

SourceClaude Opus 4.8

Claude Opus 4.8

winner

73.0

Rank #1/49 · confidence 4 · eval date 2026-05-28

+8.1

GPT-5.4

64.9

Rank #4/49 · confidence 4 · eval date 2026-03-05

Agentsreported

ResearchClawBench

SourceClaude Opus 4.8

Claude Opus 4.8

winner

21.1

Rank #1/18 · confidence 4 · eval date 2026-05-28

+5.8

GPT-5.4

15.3

Rank #13/18 · confidence 4 · eval date 2026-03-05

Agentsreported

Toolathlon

SourceClaude Opus 4.8

Claude Opus 4.8

winner

59.9

Rank #2/24 · confidence 4 · eval date 2026-05-28

+5.3

GPT-5.4

54.6

Rank #6/24 · confidence 4 · eval date 2026-03-05

Agentsreported

τ²-bench results

SourceClaude Opus 4.8

Claude Opus 4.8

94.4

Rank #22/82 · confidence 4 · eval date 2026-05-28

+4.5

GPT-5.4

winner

98.9

Rank #2/82 · confidence 4 · eval date 2026-03-05

Agentsreported

DeepSearchQA

SourceClaude Opus 4.8

Claude Opus 4.8

winner

93.1

Rank #3/14 · confidence 4 · eval date 2026-05-28

+19.5

GPT-5.4

73.6

Rank #12/14 · confidence 4 · eval date 2026-03-05

Agentsreported

OSWorld-Verified

SourceClaude Opus 4.8

Claude Opus 4.8

winner

83.4

Rank #4/27 · confidence 4 · eval date 2026-05-28

+8.4

GPT-5.4

75.0

Rank #11/27 · confidence 4 · eval date 2026-03-05

Agentsreported

MCP Atlas

SourceClaude Opus 4.8

Claude Opus 4.8

winner

82.2

Rank #5/31 · confidence 4 · eval date 2026-05-28

+11.6

GPT-5.4

70.6

Rank #19/31 · confidence 4 · eval date 2026-03-05

Agentsreported

BrowseComp

SourceClaude Opus 4.8

Claude Opus 4.8

winner

84.3

Rank #10/32 · confidence 4 · eval date 2026-05-28

+1.6

GPT-5.4

82.7

Rank #16/32 · confidence 4 · eval date 2026-03-05

Agentsreported

GDPval-AA

SourceClaude Opus 4.8

Claude Opus 4.8

winner

1593.0

Rank #11/72 · confidence 4 · eval date 2026-05-28

+286.0

GPT-5.4

1307.0

Rank #26/72 · confidence 4 · eval date 2026-03-05

Agentsreported

Terminal-Bench 2.0

SourceClaude Opus 4.8

Claude Opus 4.8

74.6

Rank #14/53 · confidence 4 · eval date 2026-05-28

+0.5

GPT-5.4

winner

75.1

Rank #13/53 · confidence 4 · eval date 2026-03-05

Agentsreported

AA Agentic Index

SourceClaude Opus 4.8

Claude Opus 4.8

42.6

Rank #18/70 · confidence 4 · eval date 2026-05-28

+1.6

GPT-5.4

winner

44.2

Rank #14/70 · confidence 4 · eval date 2026-03-05

Agentsreported

GDPval-AA

SourceClaude Opus 4.8

Claude Opus 4.8

winner

49.5

Rank #14/66 · confidence 4 · eval date 2026-05-28

+9.2

GPT-5.4

40.3

Rank #26/66 · confidence 4 · eval date 2026-03-05

Agentsreported

BenchLM Agentic score

SourceClaude Opus 4.8

Claude Opus 4.8

winner

76.7

Rank #21/76 · confidence 4 · eval date 2026-05-28

+4.7

GPT-5.4

72.0

Rank #28/76 · confidence 4 · eval date 2026-03-05

API audit comparison

Latest completed audits from shared providers, with four safety and integrity score groups plus report links.

Provider
Claude Opus 4.8
GPT-5.4
No completed audits are available from shared providers yet.

Provider examples

Speed aggregates and input/output pricing share each provider row for real API selection and migration cost checks.

Provider
Claude Opus 4.8
GPT-5.4
6345ywz API15 tests

Claude Opus 4.8

speed / latency

N/A / N/A

input / output

No data

GPT-5.4

speed / latency

41 tok/s / 6581ms

input / output

No data

XShuLab Sub2API15 tests

Claude Opus 4.8

speed / latency

N/A / N/A

input / output

No data

GPT-5.4

speed / latency

24 tok/s / 2652ms

input / output

No data

Fengsili API10 tests

Claude Opus 4.8

speed / latency

N/A / N/A

input / output

No data

GPT-5.4

speed / latency

88 tok/s / 6075ms

input / output

No data

FreeModel10 tests

Claude Opus 4.8

speed / latency

882 tok/s / 127ms

input / output

No data

GPT-5.4

speed / latency

N/A / N/A

input / output

No data

Fusecode10 tests

Claude Opus 4.8

speed / latency

43 tok/s / 2368ms

input / output

No data

GPT-5.4

speed / latency

N/A / N/A

input / output

No data

180txt API

Claude Opus 4.8

claude-opus-4-8

speed / latency

No data

input / output

$0.750/M/$3.75/M

GPT-5.4

gpt-5.4-2026-03-05

speed / latency

No data

input / output

$0/M/$0/M

Dext API

Claude Opus 4.8

claude-opus-4-8

speed / latency

No data

input / output

$0/M/$0/M

GPT-5.4

gpt-5-4

speed / latency

No data

input / output

$0/M/$0/M

兔子API

Claude Opus 4.8

claude-opus-4-8

speed / latency

No data

input / output

$0/request

GPT-5.4

gpt-5.4-openai-compact

speed / latency

No data

input / output

$0/request

熊猫 API

Claude Opus 4.8

熊猫-特供-按量-满血逆-claude-opus-4-8

speed / latency

No data

input / output

$0.015/M/$0.076/M

GPT-5.4

熊猫-按量-gpt-5.4

speed / latency

No data

input / output

$0.0054/M/$0.043/M

KFCV50

Claude Opus 4.8

claude-opus-4-8

speed / latency

No data

input / output

$0.017/request

GPT-5.4

gpt-5.4

speed / latency

No data

input / output

$0.014/request

FAQ

Weighted outcome: Claude Opus 4.8. Benchmark capability categories carry 80%, while price, API performance, and availability carry 20%.

Why is this comparison indexable?
It has 6 verifiable comparison points, and both models have pricing or benchmark data.
Are missing metrics invented?
No. Metrics without LMSpeed data are omitted from this report.

Rankings are based on community-submitted tests and periodic health probes. Advisory only, not official data.