What it is
Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite) is Google's low-cost, low-latency model, released on 21 July 2026 alongside Gemini 3.6 Flash. Google pitches it at sub-agent tasks, document parsing and high-volume agent workflows, and recommends it, with Gemini 3.8 Flash, for new projects in place of the 2.5 models, whose access is now limited.
It accepts text, images, video, audio and PDFs, returns text, and has a 1,048,576-token context with up to 65,536 output tokens. The API price is $0.30 per million input tokens and $2.50 per million output tokens; Batch and Flex are exactly half. It is on the Gemini API free tier and runs in the Gemini app and parts of Search.
Our review
Flash-Lite is a good model sold at a price that no longer looks low. Artificial Analysis scores it 22 on its Intelligence Index, 44th of 174 models and well above the median of 13 for comparable models, yet the same page calls it particularly expensive for its class. Claude Haiku 5.5 and GPT-6 Luna, both released since September, list at $0.10 input and $0.50 output against Flash-Lite's $0.30 and $2.50.
Google's numbers show a large step from 3.1 Flash-Lite: 54% against 31% on Terminal-Bench 2.1, 1140 against 642 on GDPval-AA v2, and 72.2% against 60.1% on the long-context MRCR v2 test. It also beat the older and larger Gemini 3 Flash on SWE-Bench Pro (54.2% against 49.6%) and OSWorld-Verified (74.0% against 65.1%). These are vendor figures, and Anthropic's Haiku 5.5 table uses different benchmark versions, so the two cannot be lined up directly.
Speed is uneven. Artificial Analysis measures 455.9 output tokens a second, third of 152 models, but a 7.46-second wait for the first token against a 1.64-second median, because it reasons before answering. That suits batch extraction and background agents; it is a poor fit for a chat window where a user watches a spinner.
What Flash-Lite keeps over the cheaper rivals is input breadth and Google's plumbing. It reads audio, video and PDFs natively, where Haiku 5.5 and GPT-6 Luna take text and images, and it offers Search and Maps grounding, code execution and file search on the same endpoint. The free tier is real but paid for in another currency: Google may train on free-tier prompts and let human reviewers read them.
Where it earns its place
A data engineer at an insurer processing 200,000 claim PDFs a month
Pull structured fields from scanned claim forms
Send each PDF with a JSON schema through the Batch API at $0.15 / $1.25 per million tokens. At roughly 5,000 input and 500 output tokens a document that is about $150 plus $125 a month, and the 7-second first-token wait is irrelevant in a batch job.
A media-monitoring firm handling 1,000 hours of broadcast audio a week
Summarise and tag clips without a separate transcription step
Audio is billed at the same $0.30 per million tokens as text, so one call returns tags and a summary. Neither Haiku 5.5 nor GPT-6 Luna takes audio input, which is the main reason to accept Flash-Lite's higher price; measure tokens per hour on a sample week before budgeting.
A support lead at a 30-agent SaaS helpdesk
Draft first replies to 1 million tickets a month
At about 2,000 input and 300 output tokens a reply, that is roughly $1,350 a month at standard rates against about $350 on Haiku 5.5's list price. The difference pays for a proper side-by-side evaluation on real tickets.
Before you commit
- Setup
- An AI Studio API key gives free-tier access in minutes; linking a billing account moves the project to Tier 1, usually instantly, and stops Google using its prompts for training.
- Integrations
- Gemini API with Batch, Flex and Priority inference, Google AI Studio, Android Studio and Google Cloud. Function calling, structured outputs, caching, code execution, file search, URL context and Search and Maps grounding; computer use is in preview and the Live API is not supported.
- Data and privacy
- Paid use: prompts are not used to improve Google's products and are logged for a limited, unstated period for abuse monitoring; grounded prompts are kept 30 days. Free tier and the invite-only Contributor tier: prompts may be used for training and read by human reviewers. Apps serving users in the EEA, Switzerland or the UK must use paid services.
- Cost at scale
- 1 million replies a month at 2,000 input and 300 output tokens costs about $1,350, or $675 on Batch or Flex; cached input at $0.03 cuts the cost of a repeated system prompt by 90%. Grounding beyond 5,000 free requests a month costs $14 per 1,000.
- Limits
- 1,048,576 input and 65,536 output tokens. Per-model rate limits are shown only in AI Studio and Google says they are not guaranteed; Tier 1 has a $250 billing cap and a $10 spend limit per rolling 10 minutes. No deprecation date is listed.
What's good
- 455.9 output tokens a second on Artificial Analysis
- Reads audio, video and PDF as well as text and images
- 1M-token context; Batch and Flex at half price
- Search and Maps grounding on the same endpoint
What's not
- Three times Haiku 5.5's input price and five times its output
- 7.46 seconds to first token, slow for chat
- Free-tier prompts may be used for training
- Output capped at 65,536 tokens
Best for
Teams on Google Cloud running high-volume extraction or audio and video understanding where time to first token does not matter.
Alternatives
Gemini 3.5 Flash-Lite: common questions
- Is Gemini 3.5 Flash-Lite worth it?
- Quick to stream and broad in what it reads, but Haiku 5.5 and GPT-6 Luna now charge a third of its input price and a fifth of its output. We rate it 4.1 out of 5.
- How much does Gemini 3.5 Flash-Lite cost?
- API: $0.30 input / $2.50 output per 1M tokens; cached input $0.03; Batch and Flex $0.15 / $1.25; free tier (prompts may be used for training).
- What are the downsides of Gemini 3.5 Flash-Lite?
- Three times Haiku 5.5's input price and five times its output; 7.46 seconds to first token, slow for chat; free-tier prompts may be used for training; output capped at 65,536 tokens.
- Who is Gemini 3.5 Flash-Lite best for?
- Teams on Google Cloud running high-volume extraction or audio and video understanding where time to first token does not matter.
- How much does Gemini 3.5 Flash-Lite cost for a team or in production?
- 1 million replies a month at 2,000 input and 300 output tokens costs about $1,350, or $675 on Batch or Flex; cached input at $0.03 cuts the cost of a repeated system prompt by 90%. Grounding beyond 5,000 free requests a month costs $14 per 1,000.
- Is Gemini 3.5 Flash-Lite safe for company data?
- Paid use: prompts are not used to improve Google's products and are logged for a limited, unstated period for abuse monitoring; grounded prompts are kept 30 days. Free tier and the invite-only Contributor tier: prompts may be used for training and read by human reviewers. Apps serving users in the EEA, Switzerland or the UK must use paid services.
- What are the best Gemini 3.5 Flash-Lite alternatives?
- The closest alternatives we have reviewed are Claude Haiku 5.5, GPT-6 Luna and Gemini 3.8 Flash.
Answers checked Oct 22, 2026.
Make Gemini 3.5 Flash-Lite? Show this rating on your site
