TheRouter.aiTheRouter.ai
Back to Models

Gemini 3.5 Flash

googlegoogle/gemini-3.5-flash

Google's flash-tier model in the 3.5 line.

Gemini 3.5 announcement — blog.google ↗Google I/O 2026 overview ↗

Gemini 3.5 Flash is Google's first model in the 3.5 series, announced at Google I/O on May 19, 2026. It is positioned as the Flash tier's most capable release to date — one that surpasses Gemini 3.1 Pro on key agentic and coding benchmarks while maintaining Flash-class speed and cost. Google describes it as the foundation of their 'frontier intelligence with action' strategy: a model that plans, calls tools, iterates across long-horizon tasks, and integrates with Google's Antigravity orchestration harness.

For TheRouter, Gemini 3.5 Flash is the highest-ceiling Google option currently available for agentic coding and multi-step workflows. It supports text and image input with text output, function calling, structured outputs, and OpenAI-compatible chat completions. The native pricing on the Gemini API is $1.50/M input tokens and $9.00/M output tokens (with $0.15/M for cached input); TheRouter pricing is derived from standard-models.yaml. Its 1M-token input window and dynamic thinking mode — which auto-allocates compute for harder problems — make it the default escalation path when Gemini 2.5-class models reach their quality ceiling.

Best for
  • • Agentic coding tasks that span multiple files, repositories, or multi-turn tool-calling loops — 3.5 Flash beats Gemini 3.1 Pro on Terminal-Bench 2.1 and MCP Atlas
  • • Long-horizon agentic workflows where planning, tool use, and iterative execution replace single-turn generation
  • • High-throughput multimodal tasks (text + image) needing frontier reasoning at Flash latency — 4x faster output than comparable frontier models
  • • Enterprise document reasoning over 100+ page inputs — validated by Macquarie Bank onboarding and Ramp invoice OCR use cases
Reach for something else if
  • • Ultra-cheap, high-volume classification or extraction where google/gemini-3.1-flash-lite or google/gemini-2.5-flash-lite are sufficient and meaningfully cheaper
  • • Native image or audio generation — this model outputs text only on TheRouter; use a dedicated image or audio model
  • • Tasks where Gemini 2.5 Flash quality is sufficient — prefer the lower-cost 2.5 Flash and escalate to 3.5 Flash only when benchmarks or real workloads show a quality gap
Context Length
1M
Max Output
8K
Input Priceper 1M tokens
$1.62/ 1M tokens
Output Priceper 1M tokens
$9.72/ 1M tokens

Modalities

textimage→text

Capabilities

Vision

Pricing Breakdown

TypeRate
Input$1.62 / 1M tokens
Output$9.72 / 1M tokens

Supported Parameters

temperaturemax_tokenstop_ptoolstool_choiceresponse_formatstop

Specifications

Release date2026-05-19 (Google I/O 2026)blog.google ↗verified
Knowledge cutoffJanuary 2026marktechpost.com ↗verified
Input / output modalities (native Gemini API)Text, image, audio, and video input; text output. TheRouter currently exposes text and image input with text output.marktechpost.com ↗verified
Thinking modeDynamic thinking on by default — auto-allocates more compute for harder problems without a fixed budgetmarktechpost.com ↗verified
Gemini API native pricing (input / output / cached)$1.50 / $9.00 / $0.15 per million tokens. TheRouter pricing may differ — see your dashboard.marktechpost.com ↗verified
LicenseProprietary API model; available under Google Gemini API / Google Cloud terms, not open weightsblog.google ↗verified

Benchmarks

BenchmarkDistributionScoreSource
Terminal-Bench 2.1
Exceeds Gemini 3.1 Pro on this terminal-operation and agentic coding benchmark. Reported by Google at I/O 2026.
76.2%%blog.google ↗
GDPval-AA
Real-world agentic task completion metric (Elo-based). Exceeds Gemini 3.1 Pro. Reported by Google at I/O 2026.
1656 EloEloblog.google ↗
MCP Atlas
Scaled tool-use reliability benchmark. Exceeds Gemini 3.1 Pro. Reported by Google at I/O 2026.
83.6%%blog.google ↗
CharXiv Reasoning
Multimodal understanding benchmark. Reported by Google at I/O 2026.
84.2%%blog.google ↗

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

cURL
curl https://api.therouter.ai/v1/chat/completions   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "google/gemini-3.5-flash",
    "messages": [
      {"role": "user", "content": "Summarize the key points from this input."}
    ]
  }'

API guide

Full API reference →

Chat completion

Use Gemini 3.5 Flash as an OpenAI-compatible chat model through TheRouter for agentic coding and long-context reasoning without changing SDKs.

cURL
curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"google/gemini-3.5-flash","messages":[{"role":"user","content":"Refactor this module to use async/await and add proper error handling."}]}'

More from google

google/gemini-2.5-flashPrevious-generation Flash — use when 3.5 Flash output quality is not needed and cost savings matter more.google/gemini-2.5-flash-liteLowest-cost Google option for classification, extraction, and simple high-volume tasks.google/gemini-3.1-flash-liteLightweight 3.x-series Flash for cost-optimized workloads that still want 3.x-era quality improvements.

Similar models

Cross-provider sibling models
anthropic/claude-sonnet-4.6
Comparable agentic coding default on the Anthropic side; prefer when Anthropic tool-use behavior is required.
openai/gpt-5-chat
OpenAI's frontier chat option for teams that need maximum coding accuracy over cost efficiency.
deepseek/deepseek-v3.1
Cost-focused alternative for coding and reasoning when Google or Anthropic ecosystem integration is not required.

News & changes

2026-05-19

Gemini 3.5 Flash launches at Google I/O 2026, beats Gemini 3.1 Pro on agentic benchmarks

Google unveiled Gemini 3.5 Flash at Google I/O 2026, marking the start of the 3.5 model family. The model scores 76.2% on Terminal-Bench 2.1, 83.6% on MCP Atlas, 1656 Elo on GDPval-AA, and 84.2% on CharXiv Reasoning — each exceeding Gemini 3.1 Pro — while running 4x faster on output tokens. It features dynamic thinking on by default, a 1M-token input window, and deep integration with Google's Antigravity orchestration platform.

re-authored by TheRouterblog.google ↗

Related news

Frequently asked

How is Gemini 3.5 Flash different from Gemini 2.5 Flash?

3.5 Flash surpasses Gemini 3.1 Pro on Terminal-Bench 2.1, MCP Atlas, GDPval-AA, and CharXiv Reasoning — placing it well above the 2.5 Flash class on agentic and coding benchmarks. It also runs 4x faster on output tokens than other frontier models. 2.5 Flash remains a cost-efficient option for production workloads that do not need frontier agentic ceiling.

re-authored by TheRouterblog.google ↗
Does Gemini 3.5 Flash support tool calling through TheRouter?

Yes. TheRouter exposes the model through OpenAI-compatible chat completions with tools in its supported_params. Function calling and structured output work through the standard OpenAI SDK. Antigravity-specific orchestration, Managed Agents, and native video/audio input are not available on the OpenAI-compat route.

re-authored by TheRouterblog.google ↗
What is dynamic thinking and does TheRouter expose it?

Dynamic thinking is Google's approach to adaptive reasoning compute: the model auto-allocates more deliberation for harder problems without requiring users to set a fixed thinking budget. Whether this behavior is fully exposed through TheRouter's OpenAI-compat chat completions path should be validated against the live route, as native Gemini API features vary in their OpenAI-compat translation.

re-authored by TheRoutermarktechpost.com ↗
Fact ledger — every claim on this page traces here
sourceURLretrieved
Release dateblog.google ↗2026-06-09verified
Knowledge cutoffmarktechpost.com ↗2026-06-09verified
Input / output modalities (native Gemini API)marktechpost.com ↗2026-06-09verified
Thinking modemarktechpost.com ↗2026-06-09verified
Gemini API native pricing (input / output / cached)marktechpost.com ↗2026-06-09verified
Licenseblog.google ↗2026-06-09verified
Terminal-Bench 2.1blog.google ↗2026-06-09verified
GDPval-AAblog.google ↗2026-06-09verified
MCP Atlasblog.google ↗2026-06-09verified
CharXiv Reasoningblog.google ↗2026-06-09verified
Gemini 3.5 Flash launches at Google I/O 2026, beats Gemini 3.1 Pro on agentic benchmarksblog.google ↗2026-06-09verified
How is Gemini 3.5 Flash different from Gemini 2.5 Flash?blog.google ↗2026-06-09to verify
Does Gemini 3.5 Flash support tool calling through TheRouter?blog.google ↗2026-06-09to verify
What is dynamic thinking and does TheRouter expose it?marktechpost.com ↗2026-06-09to verify
Help & contact