Gemini 3.5 Flash
Google · Released 19 May 2026 · gemini-3.5-flash
$1.50
Input price / 1M tokens
#34 of 61, cheapest first
$9.00
Output price / 1M tokens
#38 of 61, cheapest first
1.05M
Context window
#11 of 71
209
Output tokens / second
#8 of 61
76.2%
Terminal-Bench 2.1
#13 of 17
53.9%
SWE-bench Pro
#16 of 16
78.4%
OSWorld-Verified
#5 of 8
84.2%
CharXiv Reasoning
#4 of 6
Summary
Gemini 3.5 Flash is a proprietary model from Google, released on 19 May 2026.
At $1.50 input and $9.00 output per million tokens, it is #34 of 61 on input price, cheapest first.
Its 1.05M-token context window ranks #11 of 71.
Measured output speed is 209 tokens per second, #8 of 61.
Earlier Gemini 3.5 Flash model, launched at Google I/O for agentic and coding work. Google now positions it for routine, high-throughput workloads.
Benchmarks · 11 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| Terminal-Bench 2.1Terminus 2 harness | 76.2% | #13 of 17 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| SWE-bench Prosingle attempt, average of 5 runs, internal Antigravity harness | 53.9% | #16 of 16 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| MCP AtlasScale AI leaderboard | 83.6% | – | Scale AI leaderboard, cited in Google's evaluation report | |
| Toolathlonscored by the benchmark authors (HKUST) | 56.5% | – | Toolathlon authors, cited in Google's evaluation report | |
| OSWorld-Verifiedaverage of 5 runs, max 100 steps | 78.4% | #5 of 8 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| Vals Finance Agent v2 | 57.9% | – | Vals.ai, cited in Google's evaluation report | |
| CharXiv Reasoningno tools | 84.2% | #4 of 6 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| MMMU-Prono tools, average of Standard (10 options) and Vision | 83.6% | #1 of 12 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| MRCR v2 (8-needle)128k average (26.6% at 1M pointwise) | 77.3% | #4 of 13 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| Humanity’s Last Exam (no tools)full set, text + multimodal; tool setting not stated | 40.2% | #11 of 23 | Google Gemini 3.5 Flash evaluation reportSelf-reported | |
| ARC-AGI-2ARC Prize Verified, semi-private set | 72.1% | #5 of 6 | ARC Prize leaderboard, cited in Google's evaluation report |
Self-reported means the lab ran the test itself. A rank appears only where several labs report the same version of a benchmark.
Similar models
Models with the most benchmarks in common with Gemini 3.5 Flash, and the closest scores.
Terminal-Bench 2.1
#13 of 17Terminal tasks, earlier set
SWE-bench Pro
#16 of 16Harder multi-file engineering tasks
- Gemini 3.5 Flash53.9
- Gemini 3.1 Pro54.2
- Gemini 3.6 Flash58.7
- Gemini 3.5 Flash-Lite54.2
- GPT-5.558.6
- Qwen3.8-27B61.7
OSWorld-Verified
#5 of 8Using a desktop computer
CharXiv Reasoning
#4 of 6Reading charts from research papers
Compare Gemini 3.5 Flash side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- price, cache, batch, free tier: ai.google.dev/gemini-api/docs/pricing, checked 2 Oct 2026
- context, max output, modalities, api id: ai.google.dev/gemini-api/docs/models/gemini-3.5-flash, checked 2 Oct 2026
- release date, GA status: ai.google.dev/gemini-api/docs/changelog, checked 2 Oct 2026
- model card: deepmind.google/models/model-cards/gemini-3-5-flash/, checked 2 Oct 2026
- benchmarks: storage.googleapis.com/deepmind-media/gemini/gemini_3-5_flash_model_evaluation.p, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/gemini-3-5-flash, checked 2 Oct 2026
Building on Gemini 3.5 Flash?
Get the architecture right before the bill arrives.
We size caching, routing and fallbacks for your workload on a free call, and tell you where a cheaper model is good enough.