New Sign up free, 10 calls on us. Up to $1, no card needed.

Gemini 3.5 Flash-Lite

Released 2026-07-21

chatVisionCodeTool callingPrompt cachingReasoning

Gemini 3.5 Flash-Lite is a generally available, low-latency multimodal model that Google's documentation describes as the fastest, lowest-cost model in the 3.5 family, optimized for high-throughput, low-cost execution such as subagent tasks and document parsing.

Input
text image video audio $0.3/M
Output
text $2.5/M
Audio input
$0.3/M
Cache read
$0.03/M
Context
1M
vs GPT-4o
~94% cheaper
Knowledge cutoff
2026-03

Benchmarks

Above averageNo peer higher3 / 71 / 7
Gemini 3.5 Flash-Lite other models measured peer average no peer scored higher
SWE-Bench Pro
54.2%
OSWorld-Verified
74%
GDPval-AA v2 Elo · 642-1171 per Google · 2026-08-26
1140
GDM-MRCR v2 (8-needle) 128k (average)
no peer scored higher 72.2%
CharXiv (RQ) no tools
74.5%

Vendor-published: Alibaba (Qwen) Anthropic ByteDance Google Moonshot OpenAI Tencent Z.ai

Price in context

Where the price sits among 65 comparable models

Input$0.3/M
$0.05 · Qwen3 VL Flash GPT-5.4 Pro · $30
Output$2.5/M
$0.275 · DeepSeek V4 Flash GPT-5.4 Pro · $180
Cached read$0.03/M
$0.0028 · DeepSeek V4 Flash GPT-5.4 Pro · $15

The bar shows how this model’s price compares with every other model of the same kind on Synthorai. The cheapest and the most expensive are named at each end. These are base rates; batch, region and cache-write discounts are on the pricing page.

Specs & limits

Tokens

Context window (vendor spec) 1,048,576
Max output (vendor spec) 65,536
Knowledge cutoff 2026-03

Prompt caching

How it caches automatic + explicit
Min prefix 4,096

Thinking

Vendor control thinkingLevel
Accepted values minimal · low · medium · high
Default minimal applied when the request sets nothing
Can be turned off No
Thinking behaviour Defaults to the cheapest level, but minimal is a floor rather than an off switch, and Google states Flash-Lite does not support full thinking-off.
Parameter reasoning_effort
Values minimal · low · medium · high the gateway's parameter surface - the vendor mapping above applies

Model

Modalities text + image + video + audio → text
  • GA
  • 1,048,576-token context, 65,536 max output
  • the fastest, lowest-cost model in the 3.5 family, optimized for high-throughput low-cost execution (subagent tasks, document parsing)
  • thinking and built-in tools incl. Computer Use. Note: temperature/top_p/top_k are deprecated and ignored from this generation. Knowledge cutoff comes from the DeepMind model card (the API page dropped the row): March 2026, and the card adds that some domains remain limited to January 2025

per Google official docs ↗

One prompt, measured through the gateway

PROMPT Explain to a CFO, in exactly three sentences, why our LLM bill doubled after we enabled reasoning. CHECK exactly 3 sentences

Gemini 3.5 Flash-Lite passed · 3 sentences

Enabling reasoning forces the model to generate extensive, hidden "chain-of-thought" tokens before answering, effectively multiplying your input and output volume. Because LLM pricing is strictly usage-based per token, processing these invisible reasoning steps alongside your final answers instantly doubled your API costs. To control this spend, we must either restrict reasoning to complex queries or optimize our prompt architecture to reduce token bloat.

out 81 tok latency 1.9 s

Instruction following (exactly three sentences - countable), audience shaping (CFO register), and the hidden-thinking billing gap the token meters below expose.

PROMPT This function is supposed to return the longest run of consecutive equal items, but callers report it is off by one on single-element inputs and misses runs that end at the last index. Fix it and explain the bug in one paragraph. CHECK fix passes tests

Gemini 3.5 Flash-Lite passed · 8/8 cases

Here is the corrected function: ```python def longest_run(xs): if not xs: return 0 best = 1 cur = 1 for i in range(1, len(xs)): if xs[i] == xs[i - 1]: cur += 1 else: cur = 1 best = max(best, cur) return best ``` ### Explanation The original function suffered from two main counting issues: it initialized `best` and `cur` to `0` instead of `1` (causing a single-element list like `[1]` to return `0` instead of `1`), and it only updated `best` inside the `if` block when a match was found. Because the `max()` update was skipped in the `else` block and after the loop terminated, any run that ended at the final index or occurred as a single isolated element was never captured by `best`. […]

out 251 tok latency 2.6 s

Whether the fix is actually correct (runnable), explanation density, and token efficiency on a bounded task.

PROMPT Extract a JSON object with fields {company, quarter, revenue_usd_m, growth_pct, guidance} from this text. Use null for anything not stated; add no other fields. CHECK valid JSON, schema exact

Gemini 3.5 Flash-Lite passed · 5/5 fields, guidance null

```json { "company": "Meridian Robotics", "quarter": "September", "revenue_usd_m": 412, "growth_pct": 11, "guidance": null } ```

out 58 tok latency 1.0 s

Schema obedience (no invented fields), hallucination pressure (guidance is explicitly withheld), and structured-output path differences.

PROMPT Write a 120-word product announcement for an API gateway feature that caches prompts across providers. Forbidden words: "seamless", "unlock", "game-changer", "revolutionize", "empower". Exactly one sentence must be a question. CHECK 120 words, 0 banned words

Gemini 3.5 Flash-Lite passed · 115 words, 0 banned, 1 question

Managing multiple LLM providers just got faster and cheaper. Our API gateway now features intelligent prompt caching that spans across OpenAI, Anthropic, and other major AI models. By storing and reusing frequent responses at the edge, this update slashes latency and significantly reduces your overall token costs. You configure the cache rules once, and our routing engine handles the rest behind the scenes. Why pay twice for the exact same answer when your infrastructure can remember it? Developers can deploy this capability instantly with zero code changes to existing applications. Stop wasting valuable compute cycles on duplicate requests today. Upgrade your routing pipeline and experience immediate performance gains across every integrated model you currently use.

out 130 tok latency 2.0 s

Constraint obedience (word budget, banned-word list, the single question), style fingerprint, and length control.

Use Gemini 3.5 Flash-Lite in 30 seconds

OpenAI-compatible: swap the base_url, keep your SDK. POST /v1/chat/completions

from openai import OpenAI

client = OpenAI(
    base_url="https://synthorai.io/v1",
    api_key="sk-syn-...",
)

resp = client.chat.completions.create(
    model="gemini-3.5-flash-lite",
    messages=[{"role": "user", "content": "Summarize this diff"}],
)
print(resp.choices[0].message.content)

About Gemini 3.5 Flash-Lite

  • Google's card adds translation and classification to that list and frames it as the model to reach for when high throughput is critical, including agentic search.
  • It takes text, image, video, and audio input across a 1M-token context window with 65,536 output tokens, and supports function calling, structured outputs, code execution, Search and Maps grounding, URL context, file search, context caching, the Batch API, Flex and Priority inference, and computer use in preview.
  • Thinking uses thinking_level with minimal, low, medium, and high, defaulting to minimal, the cheapest default in the Gemini 3 line, though minimal is a floor rather than an off switch, so every call carries thinking tokens billed at the output rate.
  • The knowledge cutoff is March 2026, with Google noting some domains remain limited to January 2025.
  • The model page also flags an API change: temperature, top_p, and top_k are deprecated and currently ignored, and Google says future generations will reject them with an HTTP 400, recommending explicit system instructions in their place.
  • Synthorai serves it through its OpenAI-compatible chat endpoint.

FAQ

Is the Gemini 3.5 Flash-Lite API free to try?

Yes: new accounts get 10 trial calls and up to $1 in free credit, no card required. At $0.3/M input tokens, that credit alone covers roughly 416 requests of ~8K tokens against Gemini 3.5 Flash-Lite.

What is Gemini 3.5 Flash-Lite best at?

Fastest, lowest-cost model in the 3.5 family; built for high-throughput subagents and parsing; 1M context, thinking, and built-in tools. See the About section for the full picture from the vendor's own release notes.

How much does Gemini 3.5 Flash-Lite cost?

Gemini 3.5 Flash-Lite costs $0.3 per million input tokens and $2.5 per million output tokens on Synthorai. That is the provider's list price, with no platform markup. Cached input tokens bill at $0.03/M.

Does Gemini 3.5 Flash-Lite support prompt caching?

Yes: automatic caching is on by default, with an explicit mode for guaranteed savings. Cached input tokens bill at $0.03/M vs $0.3/M uncached; prompts need a 4,096-token stable prefix to cache. Prompt caching guide →

How do I get access to Gemini 3.5 Flash-Lite?

Point your existing OpenAI SDK at base_url="https://synthorai.io/v1", set model="gemini-3.5-flash-lite", and you're done. One API key covers every model on the gateway.

What is Gemini 3.5 Flash-Lite's knowledge cutoff?

Gemini 3.5 Flash-Lite's knowledge cutoff is 2026-03, per the vendor's official documentation (as of 2026-07-22).

Related models

Compare

Every value on this page is transcribed from the vendor's own documentation, linked above, and carries the date it was checked. Prices are compared across the catalogue; specification values that vendors define differently are shown with the difference stated rather than charted. Nothing here is measured by us, and nothing is scored.

Get your API key Compare your cost →