benchmarks
Gemini 3 Flash is Google's speed-oriented reasoning model with multimodal input and text output, introduced for interactive applications, coding and agentic workflows. Its published launch evaluations show stronger reasoning and coding performance than Gemini 2.5 Flash; the results below are launch measurements, not current leaderboard rankings.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 90.4%Dec 2025 | View |
| Humanity's Last Exam (without tools) | 33.7%Dec 2025 | View |
| SWE-bench VerifiedResolving real GitHub issues | 78%Dec 2025 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 95.6%6 Aug 2026 | View |
| Terminal-BenchLong-horizon command-line agent tasks | 64.3%17 Dec 2025 | View |
| FrontierMath | 35.6%17 Dec 2025 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 66.8%27 Aug 2026 | View |
| LMArena WebDevHuman preference Elo for building web apps (not a percentage) | 1440 Elo23 Oct 2026 | View |
Weighted mean of 7 verified benchmark results: GPQA Diamond 90.4%, Humanity's Last Exam (without tools) 33.7%, SWE-bench Verified 78%, OTIS Mock AIME 2024-25 95.6%, Terminal-Bench 64.3%, FrontierMath 35.6%, SimpleQA Verified 66.8%.
Capabilities
- Accepts text, image, audio and video inputs and generates text responses.
- Supports long-context document analysis with up to 1,048,576 input tokens.
- Handles reasoning, code generation and software-engineering tasks.
- Supports function calling for application-controlled tool use.
- Supports structured outputs for integration into software workflows.
- Provides configurable thinking levels to balance reasoning effort and latency.
What improved
- Higher published reasoning performance than Gemini 2.5 Flash.
- Improved multimodal understanding relative to Gemini 2.5 Flash.
- Stronger published coding and agentic task performance than Gemini 2.5 Flash.
Known flaws & limitations
- Generated factual claims can be incorrect and require independent verification.
- Native output is text; multimodal input support does not provide native image or audio generation.
- Thinking tokens contribute to billed output, so reasoning settings can materially affect cost.
- Published benchmark results depend on evaluation settings and do not guarantee equivalent application performance.
Risks, safety & security issues
- Incorrect answers can mislead users in consequential workflows without independent checks.
- Tool-connected applications can expose sensitive data or execute unintended actions if untrusted content redirects model behavior.
- Harmful or biased generated content requires application-level safeguards and monitoring.
Good fit for
- Interactive assistants that balance reasoning capability, latency and cost.
- Coding assistance and software-engineering workflows with automated tests and human review.
- Analysis and summarization of long documents and multimodal inputs.
- Tool-assisted workflows with narrowly scoped permissions and validated outputs.
Poor fit for
- Unsupervised medical, legal or financial decisions requiring guaranteed accuracy.
- Applications requiring native image or audio generation from this model.
- Autonomous workflows with unrestricted access to sensitive data or irreversible actions.
References
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.