Claude Opus 5.5 vs Gemini 3.8 Flash
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
Opus 5.554.7
Gemini 3.8 Flash42.1
Arena Text · overall Arena Text · overall: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,508
Gemini 3.8 Flash · high1,496
Arena Text · Chinese Arena Text · chinese: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,583.6
Gemini 3.8 Flash · high1,550.8
Arena Text · English Arena Text · english: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,516.9
Gemini 3.8 Flash · high1,499
Arena Text · French Arena Text · french: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,505.3
Gemini 3.8 Flash · high1,491.7
Arena Text · instruction following Arena Text · instruction following: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,535.8
Gemini 3.8 Flash · high1,491
Arena Text · non-English Arena Text · non english: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,506.9
Gemini 3.8 Flash · high1,491.1
Arena Text · Russian Arena Text · russian: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,519.1
Gemini 3.8 Flash · high1,514.4
Arena Text · Spanish Arena Text · spanish: Human pairwise preferences in the stated text vote pool
Opus 5.5 · high1,519
Gemini 3.8 Flash · high1,485.7
AA Omniscience Index AA Omniscience Index: Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty
Opus 5.540.3
Gemini 3.8 Flash28.6
SimpleQA Verified v1 SimpleQA Verified v1: Accuracy on short factual questions with verified answers, answered without web access
Opus 5.5 · max72.2
Gemini 3.8 Flash74.6
CritPt (physics) CritPt (physics): Research-level physics problems that need many careful reasoning steps
Opus 5.527.7
Gemini 3.8 Flash12.3
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
Opus 5.551.2
Gemini 3.8 Flash39.8
AA Omniscience · Accuracy AA Omniscience · Accuracy: Measures correct answers on the AA Omniscience factual knowledge test
Opus 5.564.6
Gemini 3.8 Flash53
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
Opus 5.568.4
Gemini 3.8 Flash51.9
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)
Opus 5.584.3
Gemini 3.8 Flash84
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
Opus 5.525.6
Gemini 3.8 Flash22.8
MLCR-AA MLCR-AA: Models synthesize long medical records; the headline credits concise answers judged both accurate and complete
Opus 5.5 · max66.7
Gemini 3.8 Flash · high21.7
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
Opus 5.598.3
Gemini 3.8 Flash97.8
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
Opus 5.597.5
Gemini 3.8 Flash97.5
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
Opus 5.593.5
Gemini 3.8 Flash87.5
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
Opus 5.587.5
Gemini 3.8 Flash82.9
Chess Puzzles Chess Puzzles: Models choose the best next move from a novel chess position
Opus 5.5 · max67
Gemini 3.8 Flash · high58
FrontierMath Erdos FrontierMath Erdos: Models solve expert-written advanced mathematics problems on the named tier and edition
Opus 5.5 · max2.9
Gemini 3.8 Flash · high0
FrontierMath Tier 4 (v2) FrontierMath Tier 4 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition
Opus 5.5 · max95
Gemini 3.8 Flash · high22
FrontierMath Tiers 1–3 (v2) FrontierMath Tiers 1–3 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition
Opus 5.5 · max91.2
Gemini 3.8 Flash · high68.4
Mystery Game Puzzles Mystery Game Puzzles: Models choose the best move in puzzle positions from an undisclosed game
Opus 5.5 · max71
Gemini 3.8 Flash · high47
MathArena · ArXivMath · August 2026 MathArena · arxiv · 08/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high86
Gemini 3.8 Flash40.4
MathArena · ArXivMath · June 2026 MathArena · arxiv · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high83.3
Gemini 3.8 Flash67.4
MathArena · ArXivMath · May 2026 MathArena · arxiv · 05/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high81.7
Gemini 3.8 Flash58.3
MathArena · ArXivMath · September 2026 MathArena · arxiv · 09/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high82.1
Gemini 3.8 Flash50
MathArena · BrokenArXiv · August 2026 MathArena · arxiv_false · 08/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high77.4
Gemini 3.8 Flash27.4
MathArena · BrokenArXiv · June 2026 MathArena · arxiv_false · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high100
Gemini 3.8 Flash20.8
MathArena · BrokenArXiv · May 2026 MathArena · arxiv_false · 05/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high95
Gemini 3.8 Flash19
MathArena · BrokenArXiv · September 2026 MathArena · arxiv_false · 09/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high76.7
Gemini 3.8 Flash26.7
Vals MedCode Vals MedCode: Medical coding accuracy on patient encounters
Opus 5.549.8
Gemini 3.8 Flash48.1
Vals MedScribe Vals MedScribe: Accuracy of clinical notes produced from consultation transcripts
Opus 5.591.4
Gemini 3.8 Flash84.5
Vals Legal Research Bench Vals Legal Research Bench: Grounded legal research; a task passes only if all criteria pass
Opus 5.550.5
Gemini 3.8 Flash38.9
Vals Science Index Vals Science Index: Scientific problem solving across the index’s component evaluations
Opus 5.558.7
Gemini 3.8 Flash35.8
Arena Text · German Arena Text · german: Human pairwise preferences in the stated text vote pool
Opus 5.5—
Gemini 3.8 Flash · high1,497.6
Arena Text · Japanese Arena Text · japanese: Human pairwise preferences in the stated text vote pool
Opus 5.5—
Gemini 3.8 Flash · high1,506.3
Arena Text · Korean Arena Text · korean: Human pairwise preferences in the stated text vote pool
Opus 5.5—
Gemini 3.8 Flash · high1,461.6
Arena Text · Polish Arena Text · polish: Human pairwise preferences in the stated text vote pool
Opus 5.5—
Gemini 3.8 Flash · high1,498.5
ARC-AGI-3 · semi-private ARC-AGI-3 · semi-private: Measures interactive game solving through relative human action efficiency
Opus 5.5—
Gemini 3.8 Flash4
MathArena · ArXivMath · April 2026 MathArena · arxiv · 04/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5—
Gemini 3.8 Flash64.2
MathArena · BrokenArXiv · April 2026 MathArena · arxiv_false · 04/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5—
Gemini 3.8 Flash33.6
MathArena · ArXivLean · June 2026 MathArena · arxivlean · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5—
Gemini 3.8 Flash19.6
MathArena · ArXivLean · September 2026 MathArena · arxivlean · 09/2026: Answer accuracy on the named mathematics competition under the recorded effort
Opus 5.5 · high47.2
Gemini 3.8 Flash—
Vals LegalBench Vals LegalBench: Legal reasoning and legal-language understanding on LegalBench
Opus 5.5—
Gemini 3.8 Flash87
| Benchmark | Opus 5.5 | Gemini 3.8 Flash |
|---|---|---|
| Humanity's Last Exam | 54.7 | 42.1 |
| Arena Text · overall | 1,508 · high | 1,496 · high |
| Arena Text · Chinese | 1,583.6 · high | 1,550.8 · high |
| Arena Text · English | 1,516.9 · high | 1,499 · high |
| Arena Text · French | 1,505.3 · high | 1,491.7 · high |
| Arena Text · instruction following | 1,535.8 · high | 1,491 · high |
| Arena Text · non-English | 1,506.9 · high | 1,491.1 · high |
| Arena Text · Russian | 1,519.1 · high | 1,514.4 · high |
| Arena Text · Spanish | 1,519 · high | 1,485.7 · high |
| AA Omniscience Index | 40.3 | 28.6 |
| SimpleQA Verified v1 | 72.2 · max | 74.6 |
| CritPt (physics) | 27.7 | 12.3 |
| AA Intelligence Index 4.3.2 | 51.2 | 39.8 |
| AA Omniscience · Accuracy | 64.6 | 53 |
| AA Omniscience · Hallucination Rate | 68.4 | 51.9 |
| AA-LCR (long context) 1.1 | 84.3 | 84 |
| GDP.pdf · AA all-pass | 25.6 | 22.8 |
| MLCR-AA | 66.7 · max | 21.7 · high |
| ARC-AGI-1 · public evaluation | 98.3 | 97.8 |
| ARC-AGI-1 · semi-private | 97.5 | 97.5 |
| ARC-AGI-2 · public evaluation | 93.5 | 87.5 |
| ARC-AGI-2 · semi-private | 87.5 | 82.9 |
| Chess Puzzles | 67 · max | 58 · high |
| FrontierMath Erdos | 2.9 · max | 0 · high |
| FrontierMath Tier 4 (v2) | 95 · max | 22 · high |
| FrontierMath Tiers 1–3 (v2) | 91.2 · max | 68.4 · high |
| Mystery Game Puzzles | 71 · max | 47 · high |
| MathArena · ArXivMath · August 2026 | 86 · high | 40.4 |
| MathArena · ArXivMath · June 2026 | 83.3 · high | 67.4 |
| MathArena · ArXivMath · May 2026 | 81.7 · high | 58.3 |
| MathArena · ArXivMath · September 2026 | 82.1 · high | 50 |
| MathArena · BrokenArXiv · August 2026 | 77.4 · high | 27.4 |
| MathArena · BrokenArXiv · June 2026 | 100 · high | 20.8 |
| MathArena · BrokenArXiv · May 2026 | 95 · high | 19 |
| MathArena · BrokenArXiv · September 2026 | 76.7 · high | 26.7 |
| Vals MedCode | 49.8 | 48.1 |
| Vals MedScribe | 91.4 | 84.5 |
| Vals Legal Research Bench | 50.5 | 38.9 |
| Vals Science Index | 58.7 | 35.8 |
| Arena Text · German | — | 1,497.6 · high |
| Arena Text · Japanese | — | 1,506.3 · high |
| Arena Text · Korean | — | 1,461.6 · high |
| Arena Text · Polish | — | 1,498.5 · high |
| ARC-AGI-3 · semi-private | — | 4 |
| MathArena · ArXivMath · April 2026 | — | 64.2 |
| MathArena · BrokenArXiv · April 2026 | — | 33.6 |
| MathArena · ArXivLean · June 2026 | — | 19.6 |
| MathArena · ArXivLean · September 2026 | 47.2 · high | — |
| Vals LegalBench | — | 87 |
Specs
| Spec | Opus 5.5 | Gemini 3.8 Flash |
|---|---|---|
| Input, per 1M | $4 | !$0.75 |
| Output, per 1M | $20 | !$3.75 |
| Cost per task? | $1.34 | $0.93 |
| Speed, tok/s | 81 | — |
| Context | 1M | 1M |
| Weights | Closed | Closed |
| Released | Sep 2026 | Sep 2026 |
| See images | ✓ | ✓ |
| Read PDFs | ✓ | ✓ |
| 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
Opus 5.5 → Gemini 3.8 Flash
Sampling parameters are ignored›
temperature, top_p and top_k are ignored now and will cause HTTP 400 in future generations; remove them from requests.
No minimal thinking level›
The minimal thinking level returns an error; avoid minimal.
Gemini 3.8 Flash → Opus 5.5
No forced tool use›
Forced tool_choice any or tool returns HTTP 400; remove the forced choice.
Thinking is always on›
Disabled thinking or a manual budget returns HTTP 400; omit thinking or use adaptive thinking.
Remove sampling parameters›
Non-default temperature, top_p or top_k returns HTTP 400; omit these parameters and use prompting.
No assistant prefill›
Assistant prefill returns HTTP 400; use structured outputs or system instructions, with system instructions on Bedrock.
Old computer tool rejected · Claude API / Google Cloud›
The computer_20251124 tool returns HTTP 400 on the Claude API and Google Cloud; use computer_toolset_20260801 there.
Return thinking blocks unmodified›
Edited, reordered or partially dropped thinking blocks return HTTP 400; echo assistant messages complete and unmodified with tool results.