Google Gemini 3.5 Flash
Google unveiled Gemini 3.5 Flash at I/O 2026 as the first public model of the Gemini 3.5 family, pitched as frontier-level intelligence built for agentic workflows and coding at high speed and low cost. It claimed it outperforms the prior Gemini 3.1 Pro across most of the agentic and coding benchmark suite (e.g., Terminal-Bench 76.2% vs 70.3%, MCP tool-use 83.6% vs 78.2%) while running roughly 4x faster and about 40% cheaper. A larger Gemini 3.5 Pro was previewed at the same event but was not yet generally available, with Google saying it would arrive the following month.
Gemini 3.5 Flash became available on May 19, 2026 as the default model in the Gemini app and AI Mode in Google Search (free to consumers), and as a paid model via the Gemini API in Google AI Studio and tooling such as Antigravity, priced at $1.50 per million input tokens and $9.00 per million output tokens. Gemini 3.5 Pro remained announced-only at launch.
Five weeks in, the launch claims hold up where Google scoped them and fall short where the marketing implied 'frontier.' Independent evals confirm Google's published benchmarks closely, and the speed and agentic-tool story is real: category-leading tool-orchestration scores and very high throughput that genuinely beat the prior Gemini 3.1 Pro on coding and agentic tasks. But it sits around #8 on the Artificial Analysis Intelligence Index (a few points behind GPT-5.5 and Claude Opus 4.7), trails those rivals on hard abstract reasoning and long-context retrieval, and carries a still-elevated hallucination rate — so it's a fast, cheap agentic workhorse, not a reasoning-frontier leader. Practitioners also note the 'Flash' label now masks mid-premium pricing (input tripled versus the prior Flash) and a quietly lowered default thinking level. Most important: the larger Gemini 3.5 Pro that Google promised 'next month' had NOT reached general availability as of late June — it remains in limited preview, so the genuinely frontier-class member of this family is still not in anyone's hands.
For most executive use the fast/cheap agentic tier is now good enough to automate real tool-driven workflows — but the Gemini model that would actually top the leaderboards is still unreleased, so the 'frontier' framing is a promise, not a shipped product to plan around yet.