GEMINI 3.1 FLASH LITE
Stable Flash-Lite tier. 1M context. Native multimodal. googleSearch grounding. $0.25/$1.50 per million tokens — the cheapest frontier-grade model in the lineup.
Gemini 3.1 Flash Lite is Google's cheapest production frontier model and a workhorse budget pick in Council AI. It keeps the 1M-token context window, the multimodal stack (text + image + audio + video), and the googleSearch grounding — but trims reasoning depth and latency to hit $0.25/$1.50 per million tokens. The 3.1 designation marks it as the stable iteration of the Flash-Lite line, with better instruction-following and tool use than the original 3.0 Flash-Lite. The case for: high-throughput tasks, classification, bulk summarization, budget councils, and anywhere you'd otherwise reach for Haiku 4.5 but want native multimodal. The case against: anything requiring deeper reasoning chains, complex coding, or refined writing — promote to Flash 3, Pro 3, or a Claude model.
Use Flash Lite 3.1 for any task where the cost difference matters and the quality bar is moderate — budget councils, classification, bulk summarization, simple chat. Promote to Flash 3 ($0.50/$3) when output quality starts mattering — production chat, vision QA, anything customer-facing. Promote to Pro 3 ($2.50/$10) when you need deep reasoning, not just throughput.
3.1 is the stable Flash-Lite iteration with improved instruction-following and tool use over the initial 3.0 release. Council AI defaults to 3.1 across all tiers including free.
For high-volume, well-scoped tasks — yes. Classification, extraction, summarization, simple chat, bulk vision. For open-ended reasoning or coding, no — use Flash 3, Pro 3, or a Claude model.
$0.25 per million input tokens and $1.50 per million output tokens. Council AI bundles it inside every plan including free; no separate Google billing.
Yes — same multimodal stack as the rest of the Gemini 3 line. Text, image, audio, and video inputs in a single API call.
DeepSeek V4 Flash is cheaper per token; Flash Lite 3.1 brings multimodal and googleSearch grounding. Pick DeepSeek for text-only bulk codegen, Flash Lite for anything involving images, audio, video, or fresh information.