Gemini 3.5 Flash
Google's newest production model, released 19 May 2026 — near-Pro intelligence at Flash-tier speed, and it beats Gemini 3.1 Pro on coding while costing 25% less.
Gemini 3.5 Flash specs at a glance
| Maker | |
| Released | 19 May 2026 (Google I/O) |
| Context window | 1,048,576 tokens |
| Max output | 65,536 tokens |
| API pricing | $1.50 / $9 per 1M input / output tokens |
| Cached input | $0.15 per 1M tokens |
| Family | First public Gemini 3.5 release · 3.5 Pro rolling out separately |
| Best at | Agentic coding, high-volume production work at near-Pro quality |
Calculate API costs for this model → Compare its monthly bill against 15 other rate cards.
What's new in Gemini 3.5 Flash
Announced at Google I/O 2026 alongside the multimodal Gemini Omni, 3.5 Flash opens what Google calls its agentic Gemini era. The pitch: near-Pro intelligence at Flash-tier speed and cost. Unusually for a "Flash" model, it doesn't just approximate the Pro tier — it outperforms Gemini 3.1 Pro on coding and agentic benchmarks while undercutting it on price.
A larger Gemini 3.5 Pro — the new premium reasoning flagship, with a 2M-token context — has been announced and is rolling out separately; we'll add a page when its pricing and benchmarks are public.
Gemini 3.5 Flash benchmarks
| Benchmark | Gemini 3.5 Flash | What it measures |
|---|---|---|
| Terminal-Bench 2.1 | 76.2% | Command-line / agentic task completion |
| MCP Atlas | 83.6% | Multi-step tool orchestration |
| CharXiv Reasoning | 84.2% | Chart / figure reasoning |
| GPQA | 82.8% | Graduate-level science questions |
Key takeaway
A Flash-tier model beating its own Pro tier on coding benchmarks resets the price-performance curve: for agentic coding on a budget, 3.5 Flash is now Google's default answer — and a serious rival to Claude Sonnet 5 and DeepSeek V4.
Pricing
Gemini 3.5 Flash costs $1.50 per 1M input tokens and $9 per 1M output tokens — 25% below Gemini 3.1 Pro ($2/$12) on both sides. Cached input drops to $0.15 per 1M tokens, a 10x discount that matters for repeated-context agent loops.
Who should use Gemini 3.5 Flash
- Agentic coding on a budget — the Terminal-Bench and MCP Atlas numbers at this price are the headline.
- High-volume production — near-Pro quality with Flash economics and heavy cache discounts.
- Existing Gemini users — for most workloads it now beats 3.1 Pro at lower cost; upgrading is straightforward.
For 2M-token context or heavyweight multimodal work, Gemini 3.1 Ultra (or the incoming 3.5 Pro) is still the pick.
How Gemini 3.5 Flash compares
- Gemini 3.5 Flash vs Gemini 3.1 — should you switch? →
- Gemini 3.1 Ultra overview →
- Claude Sonnet 5 — the closest Anthropic rival →
- DeepSeek V4 — the open-weight budget option →
Frequently asked questions
When was Gemini 3.5 Flash released?
19 May 2026, announced at Google I/O. It is the first public release in the Gemini 3.5 family.
How much does Gemini 3.5 Flash cost?
$1.50 per million input tokens and $9 per million output tokens; cached input is $0.15 per million. That's 25% cheaper than Gemini 3.1 Pro.
What is its context window?
1,048,576 tokens, with a maximum output of 65,536 tokens.
How good is it at coding?
76.2% on Terminal-Bench 2.1, 83.6% on MCP Atlas and 84.2% on CharXiv Reasoning — ahead of Gemini 3.1 Pro on coding and agentic benchmarks.
Is it better than Gemini 3.1 Pro?
On coding and agentic benchmarks, yes — at 25% lower cost. Gemini 3.1 Pro/Ultra still lead on context size (2M on Ultra) and heavyweight multimodal work.