Gemini 3.8 Flash
Google's latest Flash model, launched with 3.8 Flash Cyber. Requests for Gemini 2.5 Pro are rerouted to it from 1 December 2026.
- Announced
- Access
- Public
Benchmark scores
| Test | Score | Best on the tracker | How it was measured |
|---|---|---|---|
| Epoch Capabilities Indexmany tests combined into one score | 159 points | GPT-6.2 Astra 174 points | Independentlisted as "Gemini 3.8 Flash", CI 156.59 to 160.92Run 9 Nov 2026 |
| AA Intelligence Index v5about ten tests in one score | 30.8 points | Claude Fable 5.2 64.6 points | Independentv5.0.1; listed as Gemini 3.8 Flash (High)Run 10 Nov 2026 |
| Remote Labor Indexpaid freelance projects | 5.8% | Claude Fable 5.2 41% | IndependentGemini 3.8 FlashRun 1 Oct 2026 |
| FrontierSWE V2very long engineering jobs | 20% | Claude Fable 5.2 70% | Independentproximus harness, high effort, mean@5 over 34 tasks, 20-hour budget (best@5 31.6) |
| Terminal-Bench 4.1long command-line tasks | 18% | Claude Opus 5.5 66% | Independentmini-SWE-agent harness, high effort, 330 trials, ±3.36 (95% CI) |
| APEX-Agentsprofessional office tasks | 64% | Gemini 4 Carbon 84% | Independentpass@1, unknown effort; listed as "Gemini 3.8 Flash" |
| Humanity's Last Examexpert questions, no tools | 45% | GPT-6 Astra 55% | Independentleaderboard entry "Gemini 3.8 Flash" (gemini-3.8-flash_unknown) |
| ARC-AGI-3puzzle games, standard harnessTypical first-time human player: 100% | 10% | Claude Fable 5.2 81% | IndependentARC Prize config "Gemini 3.8 Flash (High)" (google-gemini-3-8-flash-high), standard harness |
| FrontierMath T1–3hard maths problems | 68% | Claude Fable 5.2 95% | Independenthigh reasoning effort (gemini-3.8-flash_high)Run 2 Sept 2026 |
| GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated | 95% | Gemini 4 Argon 96% | Independenthigh reasoning effort (gemini-3.8-flash_high)Run 2 Sept 2026 |
| FrontierMath T4research-level maths problemsSaturated | 22% | GPT-6.1 Sol 100% | Independenthigh reasoning effort (gemini-3.8-flash_high)Run 2 Sept 2026 |
| ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated | 89% | GPT-6 Astra 95% | IndependentARC Prize config "Gemini 3.8 Flash (High)" (google-gemini-3-8-flash-high) |
| OTIS Mock AIMEcompetition mathsSaturated | 99% | GPT-5.5 100% | Independenthigh reasoning effort (gemini-3.8-flash_high)Run 2 Sept 2026 |
Sources
- Introducing Gemini 3.8 Flash and 3.8 Flash CyberGoogle DeepMind · Official · deepmind.google
- Gemini API release notesGoogle · Official · ai.google.dev
- Epoch Capabilities IndexEpoch AI · Dataset · epoch.ai
- Epoch AI Benchmarking Hub (model release dates and runs)Epoch AI · Dataset · 9 Nov 2026 · epoch.ai