gpt.college

OpenAI · GPT · Reviewed 2026-06-11

OpenAI GPT-5.4

A frontier OpenAI model release with strong reported performance across coding, tool use, multimodal, and academic reasoning benchmarks.

All models
Overall · 0–100—
Status
active
Released
—
Context
Not disclosed
Parameters
Undisclosed
Inputs
Text · Image · File
Type
multimodal
Max output
— tokens
API model ID
—
Reasoning
—
Knowledge cutoff
—
Training cutoff
—
Weights / license
—
Price / 1M tokens
—
Cached input
—
Strong SWE-Bench Pro resultHigh GPQA Diamond resultBroad multimodal and tool-use coverage
CodingReasoningResearchMultimodalAgentic

Independent benchmark scores

No snapshot score

Coding

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Reasoning

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Research

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Multimodal

—

No comparable score for this task lens in the selected independent snapshot.

No snapshot score

Agentic

—

No comparable score for this task lens in the selected independent snapshot.

Raw benchmark results

Source-backed rows with the original evaluation context preserved.

6 results
BenchmarkScoreSourceEvaluation notes
BrowseCompChecked 2026-06-11 82.7% OpenAI BrowseComp score from OpenAI's GPT-5.4 tool-use table.
FrontierMathChecked 2026-06-11 47.6% OpenAI FrontierMath Tier 1-3 score from OpenAI's GPT-5.4 academic benchmark table.
GPQA / GPQA DiamondChecked 2026-06-11 92.8% OpenAI GPQA Diamond score from OpenAI's GPT-5.4 academic benchmark table.
Humanity's Last Exam (HLE)Checked 2026-06-12 36.47% Benchmark project Scale Labs HLE Text Only score for gpt-5.4-2026-03-05 (xhigh thinking), covering the 86% text-only HLE subset.
MMMU-ProChecked 2026-06-11 81.2% OpenAI MMMU-Pro no-tools score from OpenAI's GPT-5.4 computer-use and vision table.
SWE-bench ProChecked 2026-06-11 57.7% OpenAI SWE-Bench Pro public score reported by OpenAI for GPT-5.4.

Strengths

  • Strong SWE-Bench Pro result
  • High GPQA Diamond result
  • Broad multimodal and tool-use coverage

Limitations

  • Scores are self-reported by OpenAI and should be compared with harness notes visible

Model evidence · official · Checked 2026-06-10

OpenAI models documentation

Use for current model availability and capability notes; verify before publishing claims.

Model evidence · official · Checked 2026-06-11

Introducing GPT-5.4

Official OpenAI release page with GPT-5.4 coding, tool-use, vision, and academic benchmark tables.

Task lenses preserve raw scores. Radar axes use score / current leader × 100; overall is the equal-weight average of all five indices. Missing evidence is not a zero or a low rank.