Gemini 3 Flash
Last updated: April 22, 2026
Specs
| Vendor | Google DeepMind |
|---|
| Released | December 9, 2025 |
|---|
| Context window | 1M tokens |
|---|
| Input | $0.3 / M tokens |
|---|
| Output | $2.5 / M tokens |
|---|
Benchmarks
| mmlu | 83.2 |
|---|
| gpqa | 51.8 |
|---|
| humanEval | 79.4 |
|---|
| arenaElo | 1298 |
|---|
| sweBench | 48 |
|---|
| sweBenchPro | 38 |
|---|
| liveCodeBench | 64 |
|---|
| aiderPolyglot | 62 |
|---|
| terminalBench | 32 |
|---|
| mmluPro | 73 |
|---|
| hle | 11 |
|---|
| aime | 76 |
|---|
| math500 | 89 |
|---|
| mmmu | 70 |
|---|
Strengths
- Cheapest frontier-grade multimodal
- 1M-token context
- Fast enough for voice
Weaknesses
- Reasoning gap vs Gemini 3 Pro
- Occasional refusals on edge cases
Best for
- Realtime voice and chat
- Document OCR and summarization
- Cost-sensitive RAG pipelines
Compare with other AI models