Gemini 3.5 Flash: the Flash that beats the Pro

Gemini 3.5 Flash launched at Google I/O 2026 on May 19 and broke the Flash line's usual positioning: instead of being the cheap sidekick to a Pro flagship, it beats Gemini 3.1 Pro on coding and agentic benchmarks — 76.2% on Terminal-Bench and 83.6% on MCP Atlas — while costing about 25% less ($1.50 per million input tokens and $9 per million output, vs the Pro's $2/$12). It keeps everything the Gemini line is known for: a 1M-token context window, native multimodal vision input, and googleSearch grounding for the freshest retrieval in the frontier set. Versus its Flash-line predecessor, Gemini 3 Flash ($0.50/$3), it costs 3× more but plays a different role — 3 Flash is a speed tier, 3.5 Flash is a frontier coding and agent model that happens to carry the Flash name. The case for it: agentic coding, tool-heavy workflows, and search-grounded work at a mid-tier price. The case against: tasks that lean on the Pro line's deeper reasoning, and bulk throughput where 3 Flash or 3.1 Flash-Lite remain far cheaper.

Specs

Best at

• Agentic coding — 76.2% Terminal-Bench puts it ahead of its own Pro sibling on terminal-driven tasks.
• Tool / MCP workflows — 83.6% on MCP Atlas; a natural fit for MCP-heavy agent stacks.
• Search-grounded answers — googleSearch grounding at a price where you can afford to use it constantly.
• Long-context work on a budget — 1M tokens at $1.50/M input is among the cheapest whole-repo reads available.

Where it loses

• Deep reasoning — the Pro line and the Claude/OpenAI flagships still lead on research-grade problems despite the benchmark wins on coding.
• Bulk throughput cost — Gemini 3 Flash ($0.50/$3) and 3.1 Flash-Lite ($0.25/$1.50) are far cheaper for classification-scale work.
• Refactor taste — Claude Sonnet 5 produces patches engineers merge with fewer edits.

Where it sits in the Gemini range

The Gemini lineup now brackets 3.5 Flash from both sides: Gemini 3.1 Pro ($2/$12) keeps the reasoning crown of the range, while Gemini 3 Flash ($0.50/$3) and 3.1 Flash-Lite ($0.25/$1.50) hold the throughput floor. 3.5 Flash occupies the newly interesting middle — Google's answer to 'frontier coding without flagship pricing'. If your workload is agentic or terminal-driven, start here rather than at Pro; if it's deep analysis or math, Pro still earns its premium; if it's high-volume and simple, drop down a tier.

How Council AI runs it

Gemini 3.5 Flash is available on every paid Council AI plan (Plus, Pro, Ultra) and has become a default Google seat in cost-balanced councils: frontier-adjacent quality, search grounding, and a price that doesn't crowd out other voices in your monthly budget. Councils run it in parallel with models from up to 8 other labs, and the moderator scores agreement across the answers — its googleSearch grounding often makes it the fact-checking voice when other models disagree on current events.

Frequently asked questions