Models / GPT-5.4

OpenAI

GPT-5.4

63 published results from 6 sources. Each card shows where the number comes from and what it does not measure. Results are never combined into one score.

Provider: OpenAI
Sources: 6
Our benchmarks: 0
Price: Not yet published

Reported by others

Reported by Arena (formerly LMArena)

-0.04

Confirmed task success · rank 24 of 46

Unit: IPS effect estimate, higher is better
Range: -0.05 to -0.02
Sample: 57163 observations
Configuration: GPT-5.4
Measured: 28 Sep 2026
Not shown: Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Reported by Arena (formerly LMArena)

-0.01

Praise over complaint · rank 17 of 46

Unit: IPS effect estimate, higher is better
Range: -0.04 to 0.02
Sample: 17944 observations
Configuration: GPT-5.4
Measured: 28 Sep 2026
Not shown: Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Reported by Arena (formerly LMArena)

0.02

Steerability · rank 11 of 46

Unit: IPS effect estimate, higher is better
Range: 0.01 to 0.04
Sample: 64923 observations
Configuration: GPT-5.4
Measured: 28 Sep 2026
Not shown: Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Reported by Arena (formerly LMArena)

0.00

Tool grounding · rank 1 of 46

Unit: IPS effect estimate, higher is better
Range: 0.00 to 0.00
Sample: 3672050 observations
Configuration: GPT-5.4
Measured: 28 Sep 2026
Not shown: Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Reported by Arena (formerly LMArena)

1,470

Overall · rank 11 of 44

Unit: Arena rating, higher is better
Range: 1,464 to 1,476
Sample: 33331 votes
Configuration: GPT-5.4
Measured: 13 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,195

Overall · rank 7 of 34

Unit: Arena rating, higher is better
Range: 1,189 to 1,200
Sample: 110116 votes
Configuration: GPT-5.4
Measured: 24 Aug 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,475

Overall · rank 9 of 177

Unit: Arena rating, higher is better
Range: 1,473 to 1,478
Sample: 67509 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not an error rate: claims that cannot be checked on the web are skipped, and preference still carries most of the weight.

Reported by Arena (formerly LMArena)

1,485

Overall · rank 3 of 177

Unit: Arena rating, higher is better
Range: 1,482 to 1,488
Sample: 64394 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not an error rate: claims that cannot be checked on the web are skipped, and preference still carries most of the weight.

Reported by Arena (formerly LMArena)

1,475

Business, management and finance · rank 11 of 402

Unit: Arena rating, higher is better
Range: 1,468 to 1,481
Sample: 13882 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,482

Business, management and finance · rank 8 of 402

Unit: Arena rating, higher is better
Range: 1,475 to 1,488
Sample: 13092 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,434

Creative writing · rank 36 of 407

Unit: Arena rating, higher is better
Range: 1,428 to 1,441
Sample: 11135 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,443

Creative writing · rank 25 of 407

Unit: Arena rating, higher is better
Range: 1,436 to 1,450
Sample: 10894 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,492

Expert prompts · rank 22 of 359

Unit: Arena rating, higher is better
Range: 1,484 to 1,500
Sample: 6707 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,515

Expert prompts · rank 8 of 359

Unit: Arena rating, higher is better
Range: 1,506 to 1,523
Sample: 6334 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,460

Instruction following · rank 29 of 409

Unit: Arena rating, higher is better
Range: 1,455 to 1,466
Sample: 23384 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,471

Instruction following · rank 15 of 409

Unit: Arena rating, higher is better
Range: 1,466 to 1,477
Sample: 21896 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,465

Overall · rank 38 of 409

Unit: Arena rating, higher is better
Range: 1,461 to 1,469
Sample: 67632 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,475

Overall · rank 19 of 409

Unit: Arena rating, higher is better
Range: 1,471 to 1,479
Sample: 64479 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,458

Writing, literature and language · rank 19 of 408

Unit: Arena rating, higher is better
Range: 1,452 to 1,464
Sample: 16553 votes
Configuration: GPT-5.4
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,465

Writing, literature and language · rank 15 of 408

Unit: Arena rating, higher is better
Range: 1,459 to 1,471
Sample: 16142 votes
Configuration: GPT-5.4 (high)
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Microsoft STATE-Bench

26.2%

Consistency (pass^5) · rank 4 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

38.0%

Consistency (pass^5) · rank 1 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

47.2%

Customer support task success (pass@1) · rank 3 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

57.6%

Customer support task success (pass@1) · rank 1 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

50.3%

Shopping assistant task success (pass@1) · rank 3 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

53.6%

Shopping assistant task success (pass@1) · rank 2 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

46.9%

Task success (pass@1) · rank 5 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

55.7%

Task success (pass@1) · rank 1 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

43.2%

Travel task success (pass@1) · rank 5 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

55.9%

Travel task success (pass@1) · rank 1 of 5

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

3.41

User experience · rank 3 of 5

Unit: score from 1 to 5, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not real customer satisfaction: the ratings come from a model-based judge.

Reported by Microsoft STATE-Bench

3.49

User experience · rank 1 of 5

Unit: score from 1 to 5, higher is better
Configuration: gpt-5.4
Measured: 25 May 2026
Not shown: Not real customer satisfaction: the ratings come from a model-based judge.

Reported by Microsoft STATE-Bench

41.8%

Consistency (pass^5) · rank 2 of 2

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 1 Jul 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

59.0%

Customer support task success (pass@1) · rank 1 of 2

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 1 Jul 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

55.0%

Shopping assistant task success (pass@1) · rank 1 of 2

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 1 Jul 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

58.6%

Task success (pass@1) · rank 2 of 2

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 1 Jul 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

62.0%

Travel task success (pass@1) · rank 1 of 2

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 1 Jul 2026
Not shown: Not the model alone: it runs inside STATE-Bench's agent loop against simulated users, and versions of the benchmark are not comparable.

Reported by Microsoft STATE-Bench

3.67

User experience · rank 1 of 2

Unit: score from 1 to 5, higher is better
Configuration: gpt-5.4
Measured: 1 Jul 2026
Not shown: Not real customer satisfaction: the ratings come from a model-based judge.

Reported by OpenHands Index

64.3%

Average · rank 6 of 34

Unit: % resolved, higher is better
Configuration: gpt-5.4
Measured: 30 Jun 2026
Not shown: Not the model alone: the score is for the model inside the OpenHands agent, at the SDK version shown.

Reported by OpenHands Index

36.8%

Front end (SWE-Bench Multimodal) · rank 11 of 34

Unit: % resolved, higher is better
Configuration: gpt-5.4
Measured: 30 Jun 2026
Not shown: Not the model alone: the score is for the model inside the OpenHands agent, at the SDK version shown.

Reported by OpenHands Index

56.2%

Greenfield (Commit0) · rank 3 of 34

Unit: % resolved, higher is better
Configuration: gpt-5.4
Measured: 30 Jun 2026
Not shown: Not the model alone: the score is for the model inside the OpenHands agent, at the SDK version shown.

Reported by OpenHands Index

82.4%

Information gathering (GAIA) · rank 3 of 34

Unit: % resolved, higher is better
Configuration: gpt-5.4
Measured: 30 Jun 2026
Not shown: Not the model alone: the score is for the model inside the OpenHands agent, at the SDK version shown.

Reported by OpenHands Index

75.6%

Issue resolution (SWE-Bench) · rank 10 of 34

Unit: % resolved, higher is better
Configuration: gpt-5.4
Measured: 30 Jun 2026
Not shown: Not the model alone: the score is for the model inside the OpenHands agent, at the SDK version shown.

Reported by OpenHands Index

70.4%

Testing (SWT-Bench) · rank 12 of 34

Unit: % resolved, higher is better
Configuration: gpt-5.4
Measured: 30 Jun 2026
Not shown: Not the model alone: the score is for the model inside the OpenHands agent, at the SDK version shown.

Reported by tau2-bench

21.6%

Consistency (pass^4) · rank 9 of 21

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 6 May 2026
Not shown: Not your policies or customers: the tasks and policies are tau2's own, and the customer is simulated by gpt-5.2.

Reported by tau2-bench

39.4%

Task success (pass^1) · rank 10 of 21

Unit: % of tasks, higher is better
Configuration: gpt-5.4
Measured: 6 May 2026
Not shown: Not your policies or customers: the tasks and policies are tau2's own, and the customer is simulated by gpt-5.2.

Reported by UGI Leaderboard

46.0%

Requested-length error · rank 308 of 370

Unit: % off the requested word count, lower is better
Configuration: GPT-5.4 (high)
Measured: 5 Mar 2026
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

57.0%

Requested-length error · rank 318 of 370

Unit: % off the requested word count, lower is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

53.0%

Requested-length error · rank 314 of 370

Unit: % off the requested word count, lower is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

50.0%

Requested-length error · rank 312 of 370

Unit: % off the requested word count, lower is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

30.0%

Requested-length error · rank 253 of 370

Unit: % off the requested word count, lower is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

0.40

Style adherence · rank 27 of 370

Unit: score from 0 to 1, higher is better
Configuration: GPT-5.4 (high)
Measured: 5 Mar 2026
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

0.38

Style adherence · rank 62 of 370

Unit: score from 0 to 1, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

0.39

Style adherence · rank 35 of 370

Unit: score from 0 to 1, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

0.38

Style adherence · rank 58 of 370

Unit: score from 0 to 1, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

0.40

Style adherence · rank 30 of 370

Unit: score from 0 to 1, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

68.8

Writing score · rank 33 of 370

Unit: score out of 100, higher is better
Configuration: GPT-5.4 (high)
Measured: 5 Mar 2026
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by UGI Leaderboard

64.8

Writing score · rank 66 of 370

Unit: score out of 100, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by UGI Leaderboard

67.3

Writing score · rank 46 of 370

Unit: score out of 100, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by UGI Leaderboard

58.0

Writing score · rank 112 of 370

Unit: score out of 100, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by UGI Leaderboard

68.7

Writing score · rank 34 of 370

Unit: score out of 100, higher is better
Configuration: gpt-5.4
Measured: 5 Mar 2026
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by Vectara Hallucination Leaderboard

99.9%

Answer rate · rank 19 of 108

Unit: % of documents, higher is better
Configuration: GPT-5.4
Measured: 22 Sep 2026
Not shown: Not a quality score: a low rate usually means content filters were triggered, and hallucination rates are measured on answered documents only.

Reported by Vectara Hallucination Leaderboard

7.0%

Hallucination rate · rank 32 of 108

Unit: % of summaries, lower is better
Configuration: GPT-5.4
Measured: 22 Sep 2026
Not shown: Not errors in open questions or other tasks: only summarisation, judged by Vectara's own model (HHEM-2.3), not by people, on news-style documents rather than your data.

GPT-5.4

Reported by others

Compare GPT-5.4 with