GPT-5.4 (2026-03-05) vs GPT-5.5
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
GPT-5.4 (2026-03-05)—
GPT-5.542.4
SimpleQA Verified v1 SimpleQA Verified v1: Accuracy on short factual questions with verified answers, answered without web access
GPT-5.4 (2026-03-05) · xhigh45.1
GPT-5.5 · xhigh63
Chess Puzzles Chess Puzzles: Models choose the best next move from a novel chess position
GPT-5.4 (2026-03-05)37
GPT-5.5 · xhigh63
FrontierMath Tier 4 (v2) FrontierMath Tier 4 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.4 (2026-03-05) · xhigh49
GPT-5.5 · xhigh72.5
FrontierMath Tiers 1–3 (v2) FrontierMath Tiers 1–3 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.4 (2026-03-05) · xhigh78.6
GPT-5.5 · xhigh85.3
Mystery Game Puzzles Mystery Game Puzzles: Models choose the best move in puzzle positions from an undisclosed game
GPT-5.4 (2026-03-05)28
GPT-5.5 · xhigh56
Arena Text · overall Arena Text · overall: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,481
Arena Text · Chinese Arena Text · chinese: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.51,533.2
Arena Text · English Arena Text · english: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,470.3
Arena Text · French Arena Text · french: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.51,484.7
Arena Text · German Arena Text · german: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.51,480.4
Arena Text · instruction following Arena Text · instruction following: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,478.5
Arena Text · Japanese Arena Text · japanese: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,498.5
Arena Text · Korean Arena Text · korean: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,459.3
Arena Text · non-English Arena Text · non english: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,466.9
Arena Text · Polish Arena Text · polish: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,464.4
Arena Text · Russian Arena Text · russian: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.51,472.9
Arena Text · Spanish Arena Text · spanish: Human pairwise preferences in the stated text vote pool
GPT-5.4 (2026-03-05)—
GPT-5.5 · high1,468.1
AA Omniscience Index AA Omniscience Index: Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty
GPT-5.4 (2026-03-05)—
GPT-5.518.1
CritPt (physics) CritPt (physics): Research-level physics problems that need many careful reasoning steps
GPT-5.4 (2026-03-05)—
GPT-5.518.6
AA Intelligence Index 4.3.2 AA Intelligence Index 4.3.2: Artificial Analysis combines tests of knowledge, reasoning, coding and tool use into one score
GPT-5.4 (2026-03-05)—
GPT-5.533.8
AA Omniscience · Accuracy AA Omniscience · Accuracy: Measures correct answers on the AA Omniscience factual knowledge test
GPT-5.4 (2026-03-05)—
GPT-5.556.8
AA Omniscience · Hallucination Rate AA Omniscience · Hallucination Rate: Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test · lower is better
GPT-5.4 (2026-03-05)—
GPT-5.589.6
AA-LCR (long context) 1.1 AA-LCR (long context) 1.1: Questions that need reading and connecting facts across very long documents (around 100K tokens)
GPT-5.4 (2026-03-05)—
GPT-5.583
GDP.pdf · AA all-pass GDP.pdf · AA all-pass: Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes
GPT-5.4 (2026-03-05)—
GPT-5.520.6
IFBench (instructions) IFBench (instructions): Checks whether the model follows precise formatting and content instructions
GPT-5.4 (2026-03-05)—
GPT-5.571
MLCR-AA MLCR-AA: Models synthesize long medical records; the headline credits concise answers judged both accurate and complete
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh17.8
ARC-AGI-1 · public evaluation ARC-AGI-1 · public-eval: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.4 (2026-03-05)—
GPT-5.596.5
ARC-AGI-1 · semi-private ARC-AGI-1 · semi-private: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.4 (2026-03-05)—
GPT-5.592.2
ARC-AGI-2 · public evaluation ARC-AGI-2 · public-eval: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.4 (2026-03-05)—
GPT-5.570.7
ARC-AGI-2 · semi-private ARC-AGI-2 · semi-private: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.4 (2026-03-05)—
GPT-5.570.4
ARC-AGI-3 · semi-private ARC-AGI-3 · semi-private: Measures interactive game solving through relative human action efficiency
GPT-5.4 (2026-03-05)—
GPT-5.5 · high0.4
EQ-Bench 3 EQ-Bench 3: Judge-rated responses to social and emotionally nuanced scenarios, scored through the source pairwise ranking
GPT-5.4 (2026-03-05)—
GPT-5.51,576.9
FrontierMath FrontierMath: Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.4 (2026-03-05) · xhigh47.6
GPT-5.5—
FrontierMath Erdos FrontierMath Erdos: Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh0
FrontierMath Tier 4 FrontierMath Tier 4: Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.4 (2026-03-05) · xhigh27.1
GPT-5.5—
MathArena · Apex MathArena · apex · Apex: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh80.2
MathArena · Apex Shortlist MathArena · apex · Apex Shortlist: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh98.4
MathArena · ArXivMath · April 2026 MathArena · arxiv · 04/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh68.8
MathArena · ArXivMath · February 2026 MathArena · arxiv · 02/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh73.4
MathArena · ArXivMath · January 2026 MathArena · arxiv · 01/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh73.9
MathArena · ArXivMath · June 2026 MathArena · arxiv · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.583.6
MathArena · ArXivMath · March 2026 MathArena · arxiv · 03/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh77.5
MathArena · ArXivMath · May 2026 MathArena · arxiv · 05/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.577.5
MathArena · BrokenArXiv · April 2026 MathArena · arxiv_false · 04/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh72.1
MathArena · BrokenArXiv · February 2026 MathArena · arxiv_false · 02/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh68.2
MathArena · BrokenArXiv · June 2026 MathArena · arxiv_false · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.569.4
MathArena · BrokenArXiv · March 2026 MathArena · arxiv_false · 03/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh73.7
MathArena · BrokenArXiv · May 2026 MathArena · arxiv_false · 05/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.550
MathArena · ArXivLean · March 2026 MathArena · arxivlean · 03/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh17.1
MathArena · Kangaroo 2025 1-2 MathArena · kangaroo · Kangaroo 2025 1-2: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.595.8
MathArena · Kangaroo 2025 11-12 MathArena · kangaroo · Kangaroo 2025 11-12: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.598.3
MathArena · Kangaroo 2025 3-4 MathArena · kangaroo · Kangaroo 2025 3-4: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.589.6
MathArena · Kangaroo 2025 5-6 MathArena · kangaroo · Kangaroo 2025 5-6: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.590
MathArena · Kangaroo 2025 7-8 MathArena · kangaroo · Kangaroo 2025 7-8: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.595.8
MathArena · Kangaroo 2025 9-10 MathArena · kangaroo · Kangaroo 2025 9-10: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5100
MathArena · USAMO 2026 MathArena · usamo · USAMO 2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.4 (2026-03-05)—
GPT-5.5 · xhigh98.2
Vals LegalBench Vals LegalBench: Legal reasoning and legal-language understanding on LegalBench
GPT-5.4 (2026-03-05)—
GPT-5.586.5
Vals MedCode Vals MedCode: Medical coding accuracy on patient encounters
GPT-5.4 (2026-03-05)—
GPT-5.549.1
Vals MedScribe Vals MedScribe: Accuracy of clinical notes produced from consultation transcripts
GPT-5.4 (2026-03-05)—
GPT-5.586.9
Vals Legal Research Bench Vals Legal Research Bench: Grounded legal research; a task passes only if all criteria pass
GPT-5.4 (2026-03-05)—
GPT-5.540.4
| Benchmark | GPT-5.4 (2026-03-05) | GPT-5.5 |
|---|---|---|
| Humanity's Last Exam | — | 42.4 |
| SimpleQA Verified v1 | 45.1 · xhigh | 63 · xhigh |
| Chess Puzzles | 37 | 63 · xhigh |
| FrontierMath Tier 4 (v2) | 49 · xhigh | 72.5 · xhigh |
| FrontierMath Tiers 1–3 (v2) | 78.6 · xhigh | 85.3 · xhigh |
| Mystery Game Puzzles | 28 | 56 · xhigh |
| Arena Text · overall | — | 1,481 · high |
| Arena Text · Chinese | — | 1,533.2 |
| Arena Text · English | — | 1,470.3 · high |
| Arena Text · French | — | 1,484.7 |
| Arena Text · German | — | 1,480.4 |
| Arena Text · instruction following | — | 1,478.5 · high |
| Arena Text · Japanese | — | 1,498.5 · high |
| Arena Text · Korean | — | 1,459.3 · high |
| Arena Text · non-English | — | 1,466.9 · high |
| Arena Text · Polish | — | 1,464.4 · high |
| Arena Text · Russian | — | 1,472.9 |
| Arena Text · Spanish | — | 1,468.1 · high |
| AA Omniscience Index | — | 18.1 |
| CritPt (physics) | — | 18.6 |
| AA Intelligence Index 4.3.2 | — | 33.8 |
| AA Omniscience · Accuracy | — | 56.8 |
| AA Omniscience · Hallucination Rate | — | 89.6 |
| AA-LCR (long context) 1.1 | — | 83 |
| GDP.pdf · AA all-pass | — | 20.6 |
| IFBench (instructions) | — | 71 |
| MLCR-AA | — | 17.8 · xhigh |
| ARC-AGI-1 · public evaluation | — | 96.5 |
| ARC-AGI-1 · semi-private | — | 92.2 |
| ARC-AGI-2 · public evaluation | — | 70.7 |
| ARC-AGI-2 · semi-private | — | 70.4 |
| ARC-AGI-3 · semi-private | — | 0.4 · high |
| EQ-Bench 3 | — | 1,576.9 |
| FrontierMath | 47.6 · xhigh | — |
| FrontierMath Erdos | — | 0 · xhigh |
| FrontierMath Tier 4 | 27.1 · xhigh | — |
| MathArena · Apex | — | 80.2 · xhigh |
| MathArena · Apex Shortlist | — | 98.4 · xhigh |
| MathArena · ArXivMath · April 2026 | — | 68.8 · xhigh |
| MathArena · ArXivMath · February 2026 | — | 73.4 · xhigh |
| MathArena · ArXivMath · January 2026 | — | 73.9 · xhigh |
| MathArena · ArXivMath · June 2026 | — | 83.6 |
| MathArena · ArXivMath · March 2026 | — | 77.5 · xhigh |
| MathArena · ArXivMath · May 2026 | — | 77.5 |
| MathArena · BrokenArXiv · April 2026 | — | 72.1 · xhigh |
| MathArena · BrokenArXiv · February 2026 | — | 68.2 · xhigh |
| MathArena · BrokenArXiv · June 2026 | — | 69.4 |
| MathArena · BrokenArXiv · March 2026 | — | 73.7 · xhigh |
| MathArena · BrokenArXiv · May 2026 | — | 50 |
| MathArena · ArXivLean · March 2026 | — | 17.1 · xhigh |
| MathArena · Kangaroo 2025 1-2 | — | 95.8 |
| MathArena · Kangaroo 2025 11-12 | — | 98.3 |
| MathArena · Kangaroo 2025 3-4 | — | 89.6 |
| MathArena · Kangaroo 2025 5-6 | — | 90 |
| MathArena · Kangaroo 2025 7-8 | — | 95.8 |
| MathArena · Kangaroo 2025 9-10 | — | 100 |
| MathArena · USAMO 2026 | — | 98.2 · xhigh |
| Vals LegalBench | — | 86.5 |
| Vals MedCode | — | 49.1 |
| Vals MedScribe | — | 86.9 |
| Vals Legal Research Bench | — | 40.4 |
Specs
| Spec | GPT-5.4 (2026-03-05) | GPT-5.5 |
|---|---|---|
| Input, per 1M | !$2.50 | $5 |
| Output, per 1M | !$15 | $30 |
| Cost per task? | — | $0.90 |
| Speed, tok/s | — | 89 |
| Context | 1.05M | 1.05M |
| Weights | Not verified | Closed |
| Released | Mar 2026 | Apr 2026 |
| See images | ✓ | ✓ |
| Read PDFs | ? | Partial |
| Hear audio | — | — |
| Watch video | — | — |
| Call tools | ✓ | ✓ |
| Force a tool call | ? | Partial |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
GPT-5.4 (2026-03-05) → GPT-5.5
Prompts with >272K input tokens›
Above 272K input tokens, input costs 2× and output 1.5× for the full session in Standard, Batch and Flex.
GPT-5.5 → GPT-5.4 (2026-03-05)
Nothing known.