Gemini 3.8 Flash vs GPT-6 Astra
Vals Science Index?
Gemini 3.8 Flash35.8
GPT-6 Astra65.1
| Benchmark | Gemini 3.8 Flash | GPT-6 Astra |
|---|---|---|
| Humanity's Last Exam | 42.1 | 52.7 |
| Arena Text · overall | 1,496 · high | 1,477 · max |
| Arena Text · Chinese | 1,552.5 · high | 1,489.3 · max |
| Arena Text · English | 1,498.4 · high | 1,447.6 · max |
| Arena Text · French | 1,494.8 · high | 1,424.7 · max |
| Arena Text · German | 1,509.2 · high | 1,454.7 · max |
| Arena Text · instruction following | 1,491.6 · high | 1,455.3 · max |
| Arena Text · Japanese | 1,511.3 · high | 1,403.7 · max |
| Arena Text · Korean | 1,471.1 · high | 1,427.2 · max |
| Arena Text · non-English | 1,490.9 · high | 1,432.2 · max |
| Arena Text · Polish | 1,496.1 · high | 1,427.4 · max |
| Arena Text · Russian | 1,515.8 · high | 1,441.5 · max |
| Arena Text · Spanish | 1,485 · high | 1,422.5 · max |
| AA Omniscience Index | 28.6 | 42.2 |
| SimpleQA Verified v1 | 74.6 | 75.8 |
| CritPt (physics) | 12.3 | 29.1 |
| AA Intelligence Index 4.3.2 | 39.8 | 49.6 |
| AA Omniscience · Accuracy | 53 | 60.6 |
| AA Omniscience · Hallucination Rate | 51.9 | 46.5 |
| AA-LCR (long context) 1.1 | 84 | 79.7 |
| GDP.pdf · AA all-pass | 22.8 | 30.4 |
| MLCR-AA | 12.8 | 38.3 |
| ARC-AGI-1 · public evaluation | 97.8 | 98.8 |
| ARC-AGI-1 · semi-private | 97.5 | 97.5 |
| ARC-AGI-2 · public evaluation | 87.5 | 96.7 |
| ARC-AGI-2 · semi-private | 82.9 | 92.1 |
| ARC-AGI-3 · semi-private | 4 | 38.6 |
| Chess Puzzles | 58 · high | 74 · max |
| FrontierMath Erdos | 0 · high | 2.9 · max |
| FrontierMath Tier 4 (v2) | 22 · high | 97.6 |
| FrontierMath Tiers 1–3 (v2) | 68.4 · high | 93.7 · max |
| Mystery Game Puzzles | 47 · high | 84 · max |
| MathArena · ArXivMath · April 2026 | 64.2 | 90.8 · max |
| MathArena · ArXivMath · August 2026 | 40.4 | 88.6 · max |
| MathArena · ArXivMath · June 2026 | 67.4 | 94.4 · max |
| MathArena · ArXivMath · May 2026 | 58.3 | 95 · max |
| MathArena · BrokenArXiv · April 2026 | 33.6 | 97.5 · max |
| MathArena · BrokenArXiv · August 2026 | 27.4 | 81.9 · max |
| MathArena · BrokenArXiv · June 2026 | 20.8 | 99.1 · max |
| MathArena · BrokenArXiv · May 2026 | 19 | 91 · max |
| MathArena · ArXivLean · June 2026 | 19.6 | 65.2 · max |
| Vals MedCode | 48.1 | 48.5 |
| Vals MedScribe | 84.5 | 87.9 |
| Vals Legal Research Bench | 38.9 | 39.4 |
| Vals Science Index | 35.8 | 65.1 |
| MathArena · ArXivMath · September 2026 | 50 | — |
| MathArena · BrokenArXiv · September 2026 | 26.7 | — |
| MathArena · ArXivLean · September 2026 | — | 37.7 · max |
| Vals LegalBench | 87 | — |
Specs
| Spec | Gemini 3.8 Flash | GPT-6 Astra |
|---|---|---|
| Input, per 1M | !$0.75 | $10 |
| Output, per 1M | !$3.75 | $50 |
| Cost per task? | $0.93 | $1.54 |
| Speed, tok/s | — | 59 |
| Context | 1M | 1.05M |
| Weights | Closed | Closed |
| Released | Sep 2026 | Sep 2026 |
| See images | ✓ | ✓ |
| Read PDFs | ✓ | Partial |
| Hear audio | ✓ | — |
| Watch video | ✓ | — |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | Partial |
| Return strict JSON | ✓ | ✓ |
| Search the web | ✓ | ✓ |
| Run code | ✓ | ✓ |
| Use a computer | ? | ✓ |
What breaks when you switch
Gemini 3.8 Flash → GPT-6 Astra
Monitoring can return HTTP 403›
Misalignment monitoring can return HTTP 403 before streaming or errors after output; match misalignment_policy_violation and handle stream errors.
No none reasoning effort›
The none reasoning effort is unsupported on GPT-6 Astra and GPT-6.1 Sol; use low instead.
Remove unsupported sampling parameters›
When reasoning effort is not none, remove temperature, top_p and top_logprobs, plus Chat Completions logprobs or Responses output logprobs inclusion.
GPT-6 Astra → Gemini 3.8 Flash
Context: 1.05M → 1M
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.