Gemini 3.8 Flash
64.5#65 of 427 overall
- Input / output
- US$ 0,83 / US$ 4,13 US$ per 1M tokens
- Context
- 1.0M tokens
- Release
- 09/02/2026 release date
- Weights
- Closed
- Speed
- 127 tok/s first token in 18.01 s
Family history
The 8 versions of this family, from first to newest, by release date. Click a point to open that version.
See the data as a table
| Model | Released | Score | Rank | AA Intel. | Elo Arena |
|---|---|---|---|---|---|
| Gemini 1.5 Flash | 05/14/2024 | 21.8 | #395 | — | 1287 |
| Gemini 2.0 Flash Thinking Experimental | 01/21/2025 | 34.7 | #307 | — | 1354 |
| Gemini 2.5 Flash | 06/17/2025 | 54.5 | #106 | — | 1417 |
| Gemini 3 Flash Preview | 12/17/2025 | 70.8 | #36 | — | 1466 |
| Gemini 3.5 Flash | 05/19/2026 | 75.8 | #24 | — | 1482 |
| Gemini 3.6 Flash | 07/21/2026 | 76.4 | #22 | — | 1480 |
| Gemini 3.7 Flash | 08/13/2026 | 81.8 | #13 | — | 1489 |
| Gemini 3.8 Flash | 09/02/2026 | 64.5 | #65 | 30.5 | 1498 |
Scores by source
Scores by sourcepercentile = % of ranked models it beats| Metric | Value | Percentile | Source |
|---|---|---|---|
| LiveBench weight 3 | 59.2 | 29% | LiveBench |
| AA Intelligence Index weight 3 | 30.5 | 61% | Artificial Analysis |
| Arena text (Elo) weight 2 | 1498 | 96% | LMArena |
| Arena WebDev (Elo) weight 1 | 1583 | 74% | LMArena |
| Arena vision (Elo) weight 0.5 | 1315 | 94% | LMArena |
| Humanity's Last Exam weight 1 | 47.8 | 96% | Artificial Analysis |
| Terminal-Bench weight 1 | 19.7 | 78% | Artificial Analysis |
| GPQA Diamond | 95.3 | 100% | Artificial Analysis |
| Long-context reasoning | 81.3 | 91% | Artificial Analysis |
| LiveBench long agentic | 16.5 | 23% | LiveBench |
| LiveBench reasoning | 89.3 | 61% | LiveBench |
| LiveBench coding | 82.2 | 39% | LiveBench |
| LiveBench browser | 45.1 | 36% | LiveBench |
| LiveBench data analysis | 54.0 | 7% | LiveBench |
| LiveBench research | 79.1 | 58% | LiveBench |
| LiveBench design | 72.7 | 23% | LiveBench |
By category
| Category | Score | Rank | Metrics |
|---|---|---|---|
| Agentic | 50.3 | #64 | LiveBench long agentic, Terminal-Bench |
| Coding | 54.3 | #45 | LiveBench coding, Arena WebDev (Elo) |
| Reasoning | 76.7 | #33 | LiveBench reasoning, Humanity's Last Exam, GPQA Diamond |
| Data analysis | 28.3 | #30 | LiveBench data analysis |
| Research | 66.2 | #53 | LiveBench research, Long-context reasoning |
| Design | 48.9 | #59 | LiveBench design, Arena WebDev (Elo) |
| Browser use | 42.8 | #21 | LiveBench browser |
| Chat | 73.2 | #8 | Arena text (Elo) |
| Vision | 72.2 | #7 | Arena vision (Elo) |
History
RankingIA score on each collection day (hover to see the rank).
Daily collection started on 10/30/2026. The chart appears from the second day the model is ranked.