Gemini 3.5
Google's frontier model for agents and coding, made fast and cheap.
Context window
1,048,576 tokens (Gemini 3.5 Flash; Pro variant not yet released)
Input / 1M tokens
$1.50
Output / 1M tokens
$9.00
Provider
Google DeepMind
Pricing shown is for the generally available Gemini 3.5 Flash ($1.50 input / $9.00 output per 1M tokens; cached input $0.15/1M). Gemini 3.5 Pro was announced at Google I/O 2026 but pricing is Not announced as it had not launched publicly as of July 2026. · Data verified 2026-07-02
Gemini 3.5 Flash is the first model in Google's Gemini 3.5 family, launched at Google I/O 2026 (May 19-20, 2026) and set as the default across the Gemini app, Search AI Mode, Antigravity, and the Gemini API. It pairs frontier-level intelligence with agentic action, excelling at long-horizon multi-step tasks, tool use, and coding while running ~4x faster on output tokens and at less than half the cost of prior Pro-tier models. It reportedly outperforms Gemini 3.1 Pro on several agentic and coding benchmarks. A larger Gemini 3.5 Pro variant was announced as coming soon but had not launched publicly as of July 2026.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Use via Google AI Studio or the Gemini API (get a key at aistudio.google.com), or through Vertex AI on Google Cloud. Gemini 3.5 Flash is the default model in the Gemini app, Search AI Mode, and the API.
Strengths
- ✓Agentic coding and long-horizon, multi-step problem-solving
- ✓Strong tool-use / function-calling reliability (MCP Atlas 83.6%)
- ✓Fast output generation (~4x faster on output tokens vs prior models)
- ✓Low cost relative to frontier performance
- ✓Advanced multimodal understanding (CharXiv 84.2%)
Best for developers who...
When to choose it (and when not to)
Reach for Gemini 3.5 when...
- →You need frontier agent/coding performance without frontier prices
- →Building autonomous agents that make many tool calls
- →High-throughput production workloads that were previously too costly on a Pro model
- →You want a strong default multimodal model with a 1M-token context window
Look elsewhere if...
- ✕You require the absolute top-end reasoning of a dedicated Pro / Deep Think model (3.5 Pro not yet available; consider Gemini 3.1 Pro)
- ✕You need an on-prem or open-weight model
- ✕Budget is extremely tight and quality can be sacrificed (a Flash-Lite tier is cheaper)
How to use it
- ›Set an explicit thinking level for hard tasks; keep it low for simple, high-volume calls to save latency and cost
- ›Provide tools with clear JSON schemas to exploit its strong function-calling reliability
- ›Use the large context window to include full codebases or long documents rather than pre-summarizing
- ›For agentic runs, give a clear goal plus stopping conditions so it plans multi-step actions effectively
Quickstart
Pythonfrom google import genai
client = genai.Client(api_key="YOUR_API_KEY")
response = client.models.generate_content(
model="gemini-3.5-flash",
contents="Write a Python function that reverses a linked list.",
)
print(response.text)Install with `pip install google-genai`. Set GEMINI_API_KEY or pass api_key explicitly. Get a key at aistudio.google.com.
API model id: gemini-3.5-flash
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Terminal-Bench 2.1 (coding) | 76.2% | Agentic terminal/coding tasks; outperforms Gemini 3.1 Pro |
| MCP Atlas (tool use) | 83.6% | vs 79.1% Claude Opus 4.7 and 77.8% GPT-5.5 per Google |
| CharXiv Reasoning (multimodal) | 84.2% | Chart/figure reasoning |
Source: MarkTechPost coverage of Google I/O 2026 announcement
Tools powered by Gemini 3.5
Compare Gemini 3.5
Gemini 3.5 vs Claude Opus 4.8
Anthropic - 1M ctx
Gemini 3.5 vs Claude Sonnet 5
Anthropic - 1M ctx
Gemini 3.5 vs GPT-5.6
OpenAI - 1.05M ctx
Compare Gemini 3.5 with any other model
Build a comparison →All model comparisons →