Gemini 3.5 Flash

Google documents Gemini 3.5 Flash as a stable model in the Gemini 3 family with the model code gemini-3.5-flash. It accepts text, image, video, audio, and PDF inputs and returns text output. The developer docs list a 1,048,576 token input limit, a 65,536 token output limit, January 2025 knowledge cutoff, and support for thinking, function calling, structured outputs, search grounding, Maps grounding, URL context, file search, code execution, context caching, Batch API, Flex inference, and Priority inference.

• Use it when the task needs fast agentic coding loops, multi-step workflows, multimodal inputs, or long-context reasoning.
• Do not treat it as an image, audio, Live API, or computer-use output model; the documented output is text.
• For production API work, use the stable model string gemini-3.5-flash rather than the older gemini-3-flash-preview string.

Google positions Gemini 3.5 Flash around sub-agent deployment, complex coding cycles, multi-step workflows, and long-horizon tasks. DeepMind benchmark tables emphasize agentic terminal coding, SWE-Bench Pro, MCP workflows, tool use, OSWorld-style UI control, financial analysis, multimodal chart reasoning, and long-context evaluation. Those results are useful for model screening, but teams should still run their own task-specific evals before replacing an existing production coding or agent model.

The Gemini API pricing page lists standard paid pricing at $1.50 per 1M input tokens and $9.00 per 1M output tokens, with lower Batch and Flex prices and higher Priority prices. Community threads focus heavily on whether that price still feels like a Flash-tier bargain. Use official pricing for factual cost fields, and use Reddit or forum discussion only as evidence of user concern and comparison demand.

X launch posts from Google, Google DeepMind, GitHub, and Google employees reinforce the official framing: Gemini 3.5 Flash is aimed at agentic coding, long-horizon work, tool use, cache efficiency, and broad availability. Reddit discussion is more mixed. Antigravity users report unusually fast output, more detailed artifacts, and a model that feels closer to a Pro-class coding assistant, while other threads warn that benchmark strength may not transfer to every real task and that teams should run their own evals before swapping it into production.

• Positive signal: Antigravity and coding-tool users notice speed, structured output, and stronger tool-workflow fit.
• Skeptical signal: Reddit users question benchmark-heavy marketing and report task-specific underperformance in some saved evals.
• Cost signal: Reddit users repeatedly compare the official $1.50/$9.00 standard price against older Flash models, Gemini Pro-class pricing, GPT, Claude, and Flash-Lite expectations.