Google launches Gemini 3.6 Flash and two cheaper Flash variants for agent workflows
机器之心 · wechat · 2026-07-22
Google rolled out three new Flash-family models: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber.
- Gemini 3.6 Flash targets production agent workloads, with lower token usage, faster multi-step execution, and cheaper pricing than 3.5 Flash. It reportedly cuts output tokens by 17% on Artificial Analysis and improves scores on DeepSWE, MLEBench, OSWorld-Verified, and GDPval-AAv2.
- Gemini 3.5 Flash-Lite is positioned as the fastest, lowest-cost option for high-throughput agent/search/document workflows, reaching about 350 tokens/sec in third-party testing.
- Gemini 3.5 Flash Cyber is a security-tuned variant used in CodeMender for vulnerability discovery and repair, and Google says it will be released narrowly to governments and trusted partners first.
The article also notes that Gemini 3.5 Pro is still being tested with partners and that Google has already started work on Gemini 4.
Related event: Google Unveils New Gemini Models and Starts Gemini 4 Training(150 posts)→
More from coding & agent
- Agent answered wrong on zero-click blog posts after skipping the CMS cross-check — philrox_ · 2026-07-22
- A Grok skill that scrapes 500-like X posts into SQLite and a dashboard — tinyfool · 2026-07-22
- A solo founder runs 16 AI agents to sell B2B software, and code beats prompts on guardrails — mastafied · 2026-07-22
- ChatGPT desktop MCP servers show up in /mcp but still fail in chats — Zestyclose-Driver917 · 2026-07-22
- Claude turns a Pinterest idea into an infinite rolling-paper Three.js demo — nptacek · 2026-07-22
- A new build-from-scratch guide shows how to make a Claude Code-style agent — dfinke · 2026-07-22