Gemini 3.8 Flash
Made by Google DeepMind, released 2 Sep 2026. Epoch AI lists its access as “API access”. Appears strongest at coding, knowledge and writing among the models listed here.
In short
What it does well
- Coding: Top tier (12th of 77 listed models)
- Knowledge: Top tier (7th of 43 listed models)
- Writing: Top tier (7th of 77 listed models)
Where it falls short
- Maths: Middle (23rd of 43 listed models)
What we don't know yet
- Fixing real software bugs: Not enough data
The four ratings
- Top tier
- Strong
- Middle
- Behind the leaders
- Not enough data
Ratings compare only the 77 models listed here, in quarters, best first. “Not enough data” means no result yet. It is never a low score.
- Writing
- Top tierHow often people preferred its answers to writing prompts. 1,491 on Arena's scale · Source for Gemini 3.8 Flash writing score (lmarena)
- Coding
- Top tierHow often people preferred its answers to coding questions. 1,509 on Arena's scale · Source for Gemini 3.8 Flash coding score (lmarena)
- Maths
- MiddleHow many very hard maths problems it solved. 68.4% · Source for Gemini 3.8 Flash maths score (epoch)
- Knowledge
- Top tierHow many short factual questions it answered correctly. 69.7% · Source for Gemini 3.8 Flash knowledge score (epoch)
Evidence completeness: Broad evidence, 12 of 13 tests have a result.
Where it is relatively stronger or weaker
This compares each kind of request with the model's own overall position among the 77 listed models, using Arena's votes. It is about performance on different languages, kinds of request and topics, not about political or social bias.
Relatively stronger
No category stands out as clearly stronger than its overall position.
Relatively weaker
No category stands out as clearly weaker than its overall position.
Show every category we can see
| Category | Position | Arena score | Votes | Source |
|---|---|---|---|---|
| Expert-level prompts Kind of request | 9th of 77 | 1,526 95% range 1,517 to 1,535 | 5,549 | Source for Expert-level prompts |
| Hard prompts Kind of request | 7th of 77 | 1,508 95% range 1,503 to 1,514 | 30,731 | Source for Hard prompts |
| Long prompts Kind of request | 7th of 77 | 1,509 95% range 1,503 to 1,514 | 21,802 | Source for Long prompts |
| Longer conversations Kind of request | 5th of 77 | 1,504 95% range 1,496 to 1,512 | 7,190 | Source for Longer conversations |
| Chinese Language | 7th of 76 | 1,550 95% range 1,540 to 1,561 | 3,978 | Source for Chinese |
| English prompts Language | 8th of 77 | 1,498 95% range 1,492 to 1,504 | 18,714 | Source for English prompts |
| French Language | 17th of 72 | 1,494 95% range 1,478 to 1,511 | 1,682 | Source for French |
| German Language | 4th of 55 | 1,507 95% range 1,484 to 1,529 | 760 | Source for German |
| Japanese Language | 5th of 54 | 1,508 95% range 1,486 to 1,531 | 887 | Source for Japanese |
| Korean Language | 5th of 54 | 1,477 95% range 1,455 to 1,500 | 844 | Source for Korean |
| Non-English prompts Language | 5th of 77 | 1,492 95% range 1,486 to 1,497 | 29,077 | Source for Non-English prompts |
| Russian Language | 4th of 77 | 1,516 95% range 1,508 to 1,525 | 6,256 | Source for Russian |
| Spanish Language | 12th of 70 | 1,487 95% range 1,472 to 1,502 | 1,649 | Source for Spanish |
| Entertainment, sport and media Topic | 7th of 77 | 1,481 95% range 1,474 to 1,488 | 12,638 | Source for Entertainment, sport and media |
| Law and government Topic | 10th of 76 | 1,500 95% range 1,489 to 1,510 | 4,087 | Source for Law and government |
| Medicine and health Topic | 6th of 75 | 1,506 95% range 1,495 to 1,517 | 3,353 | Source for Medicine and health |
| Science Topic | 4th of 77 | 1,523 95% range 1,516 to 1,531 | 7,897 | Source for Science |
| Software and IT Topic | 12th of 77 | 1,507 95% range 1,501 to 1,513 | 17,809 | Source for Software and IT |
Bias: what we can and can't see
Our sources do not measure political lean, stereotyping, flattery (agreeing with you too readily) or safety behaviour, so we cannot say how any model does on those. What we can show is above: where a model is relatively stronger or weaker across languages, kinds of request and topics. Also, people voting in Arena can favour answer style, such as length and formatting, and our numbers do not adjust for that. See what AI bias means and what these tests can tell you.
Best-supported use cases
This model is not among the options we recommend for any of the five tasks. That reflects the rules and the models listed, not a verdict on the model.
Where an alternative might be more suitable
- Homework and maths: rated Middle on hard maths problems. Compare it with GPT-6.1 Sol, the best-supported choice for this task.
What it is
- Maker
- Google DeepMind
- Release date (Epoch AI)
- 2 Sep 2026
- Access, as listed by Epoch AI
- API access
- Licence, as listed by Arena
- Proprietary
- Maker's country (Epoch AI)
- United States of America
- Last observation in our data
- 28 Oct 2026
- Data last changed
- 29 Oct 2026, 14:36 UTC
We have not verified an official product page for this model, so we do not link to one.
How you can use it
Proprietary. The model's files are not published, so it is generally available only through its maker or partners the maker chooses.
How you can reach it (an app, a website, a developer API or a third-party tool) and what that costs are not covered by our benchmark data. Where we could verify prices from the maker's own pages, they are shown under Price. Epoch AI lists access as “API access”. See models, apps and harnesses and how AI pricing works.
Price
Price not independently verified
See Gemini plans and pricing (checked 28 Oct 2026). Page opened and read.
We never infer subscription prices from API prices. New to “tokens” and “per million”? See how AI pricing works.
Good to remember
- Benchmarks are samples of tasks. None proves performance in every situation.
- Arena scores come from people choosing between answers. That measures preference, not correctness.
- Ratings are relative to the 77 models listed here. A different set would change them.
- Sources test models at different settings. Each result shows the setting used, and we do not adjust it.
The evidence, for the curious
Show all 13 tests with settings and sources
| Test | Result | Rating | Setting used | Source identifier | Observed | Links |
|---|---|---|---|---|---|---|
| Arena text: overall preference Arena 'text' leaderboard (Bradley-Terry rating, no style control) |
1,498 Arena score 95% range 1,494 to 1,503, 47,803 votes |
Top tier | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · text/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena text: overall preference score (lmarena) |
| Arena text: creative writing Arena 'text' leaderboard, category creative_writing |
1,491 Arena score 95% range 1,484 to 1,498, 10,581 votes |
Top tier | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · text/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena text: creative writing score (lmarena) |
| Arena text: coding Arena 'text' leaderboard, category coding |
1,509 Arena score 95% range 1,503 to 1,516, 12,376 votes |
Top tier | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · text/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena text: coding score (lmarena) |
| Arena text: maths Arena 'text' leaderboard, category math |
1,530 Arena score 95% range 1,517 to 1,542, 2,496 votes |
Top tier | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · text/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena text: maths score (lmarena) |
| Arena text: following instructions Arena 'text' leaderboard, category instruction_following |
1,493 Arena score 95% range 1,487 to 1,499, 17,179 votes |
Top tier | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · text/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena text: following instructions score (lmarena) |
| Arena text: business and management Arena 'text' leaderboard, category industry_business_and_management_and_financial_operations |
1,480 Arena score 95% range 1,473 to 1,487, 8,607 votes |
Top tier | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · text/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena text: business and management score (lmarena) |
| Arena WebDev: building web apps Arena 'webdev' leaderboard (Code Arena), category overall |
1,583 Arena score 95% range 1,576 to 1,590, 12,342 votes |
Strong | effort: High, bradley-terry rating | Gemini 3.8 Flash (High) lmarena · webdev/latest |
28 Oct 2026 | Source for Gemini 3.8 Flash Arena WebDev: building web apps score (lmarena) |
| FrontierMath Tiers 1-3 (v2) FrontierMath Tiers 1-3 v2 (private set), benchmark released 2026-06-12 |
68.4% share of problems solved standard error ±2.8 points |
Middle | effort: high, best score across scorers | gemini-3.8-flash_high epoch · frontiermath_tiers_1_3_v2.csv |
2 Sep 2026 | Source for Gemini 3.8 Flash FrontierMath Tiers 1-3 (v2) score (epoch) No public run log in the data file. |
| OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025, benchmark released 2024-12-19 |
98.9% share of problems solved standard error ±0.9 points |
Strong | effort: high, best score across scorers | gemini-3.8-flash_high epoch · otis_mock_aime_2024_2025.csv |
2 Sep 2026 | Source for Gemini 3.8 Flash OTIS Mock AIME 2024-2025 score (epoch) No public run log in the data file. |
| GPQA Diamond GPQA Diamond, benchmark released 2023-11-20 |
95.4% share of questions correct standard error ±1.4 points |
Top tier | effort: high, best score across scorers | gemini-3.8-flash_high epoch · gpqa_diamond.csv |
2 Sep 2026 | Source for Gemini 3.8 Flash GPQA Diamond score (epoch) No public run log in the data file. |
| SimpleQA Verified SimpleQA Verified, benchmark released 2025-09-09 |
69.7% share of questions correct standard error ±1.5 points |
Top tier | effort: high, best score across scorers | gemini-3.8-flash_high epoch · simpleqa_verified.csv |
2 Sep 2026 | Source for Gemini 3.8 Flash SimpleQA Verified score (epoch) No public run log in the data file. |
| SWE-bench Verified (Epoch runs) | Not enough data | |||||
| FrontierCode (reported by Cognition) FrontierCode, benchmark released 2026-06-08, aggregation Mean@5 |
41.2% mean score over 5 attempts no uncertainty figure published |
Not rated (results use different tools) | effort: medium, tool: devin-cli, mean@5 | gemini-3.8-flash_medium epoch · frontiercode_external.csv |
not provided | Source for Gemini 3.8 Flash FrontierCode (reported by Cognition) score (epoch) · Run details for FrontierCode (reported by Cognition) |
Missing evidence
Not enough data for these tests:
- SWE-bench Verified (Epoch runs)
Missing results are not low scores and do not count against the model.
Where a source gives no public run log, we cannot show the run behind the score. See data sources.
Not sure this is the right one?
Tell us what you need help with and we'll narrow the options down.