> ## Documentation Index
> Fetch the complete documentation index at: https://docs.apiyi.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Gemini 3.1 Flash Lite Goes GA: The Sweet Spot for High-Throughput Agents & Low-Latency Workloads
> Gemini 3.1 Flash Lite reached general availability on May 8, 2026. 64% faster output than 2.5 Flash, 86.9% on GPQA Diamond, priced at $0.25/$1.50 per 1M tokens. Live at APIYi via official direct connection with up to 21% off via top-up promotions.
## Key Highlights
* **Now Generally Available**: Google announced GA for Gemini 3.1 Flash Lite on May 8, 2026 (UTC+8) — production-ready
* **Model Identifier Updated**: `gemini-3.1-flash-lite-preview` → `gemini-3.1-flash-lite`. Preview users should plan migration
* **Massive Speed Boost**: **64% faster** output than 2.5 Flash (381.9 vs 232.3 tokens/sec) and **2.5× faster** time-to-first-token
* **Official Pricing Parity**: \$0.25 per 1M input / \$1.50 per 1M output — identical to Google's official rates
* **Stackable Top-up Promo**: APIYi top-up bonuses bring effective price down to **79–85% of list** (up to 21% off)
## Background
On March 3, 2026, Google launched Gemini 3.1 Flash Lite Preview, targeting the "high-throughput agents + ultra-low latency" niche. During the two-month preview, agent-heavy customers like Latitude, Cartwheel, Whering, and HubX gave strong feedback on instruction-following precision, latency, unit cost, and multimodal stability.
On **May 8, 2026 (UTC+8)**, Google formally announced **general availability**. The model identifier dropped the `-preview` suffix and became `gemini-3.1-flash-lite`. With the API contract, behavior, and billing rules now stabilized, it's safe to wire into production.
APIYi has synced GA via the official direct-connection (proxy) channel. Pricing matches Google's official rates exactly, and stacking APIYi's top-up bonus brings the effective price below list — making this one of the cheapest reliable ways to consume the Gemini 3.1 lightweight tier.
## Detailed Analysis
### What changed from Preview to GA
* Old: `gemini-3.1-flash-lite-preview`
* New: `gemini-3.1-flash-lite`
* Old name still works, migration recommended
* Frozen interface contract
* Stable rate limits and billing
* Safe for production traffic
* Higher sustained output speed
* Lower TTFT
* More stable function calling & structured output
* Batch API & Caching production-ready
* Thinking levels available in production
* Full multimodal input stable
### Performance highlights (GA benchmarks)
Per Artificial Analysis and Google's official numbers:
| Metric | Gemini 3.1 Flash Lite | Gemini 2.5 Flash | Delta |
| -------------------------------------- | --------------------------------- | ---------------- | ------------ |
| Output speed (tokens/sec) | **381.9** | 232.3 | +64% |
| Time-to-first-token | **2.5×** faster than 2.5 Flash | baseline | -60% |
| GPQA Diamond | **86.9%** | — | Tier-leading |
| MMMU Pro (multimodal reasoning) | **76.8%** | — | Tier-leading |
| Arena Elo | **1432** | — | — |
| Artificial Analysis Intelligence Index | **34** (median in price tier: 21) | — | Well above |
Across 11 benchmarks Google ran internally, Gemini 3.1 Flash Lite **beat GPT-5 mini and Claude Haiku 4.5 on 6**, at materially lower per-token cost.
### Technical specs
| Spec | Value |
| ------------------------------------- | -------------------------------- |
| Model name | `gemini-3.1-flash-lite` |
| Context window | 1,048,576 tokens (1M+) |
| Max output | 65,536 tokens (64K) |
| Input modalities | Text, image, video, audio, PDF |
| Output modality | Text |
| Knowledge cutoff | January 2025 |
| Latest update | May 2026 |
| Thinking | ✅ adjustable levels |
| Function calling | ✅ |
| Structured output | ✅ |
| Code execution | ✅ |
| File search / URL context | ✅ |
| Search Grounding / Maps Grounding | ✅ |
| Batch API / Caching / Flex / Priority | ✅ |
| Channel | APIYi official direct connection |
## Real-World Use Cases
### Recommended scenarios
* Tool calling / routing / multi-step orchestration
* High-concurrency lightweight decision nodes
* SLA-sensitive agent tasks needing stable APIs
* Tabular / form / PDF structured extraction
* Bulk content moderation, classification, tagging
* Mass log summarization & normalization
* Real-time translation & interpretation assist
* UI generation, dashboard composition
* First-touch customer support, intent detection
* Image / video understanding
* Audio transcription + key info extraction
* PDF parsing & field extraction
### Code examples
Calling the GA build via APIYi:
```python theme={null}
import openai
client = openai.OpenAI(
api_key="your-apiyi-api-key",
base_url="https://api.apiyi.com/v1"
)
response = client.chat.completions.create(
model="gemini-3.1-flash-lite", # GA, no -preview suffix
messages=[
{"role": "system", "content": "You are an efficient structured-data extraction assistant."},
{"role": "user", "content": "Extract order_id, amount, and currency from the order text. Return JSON."}
],
temperature=0.2,
response_format={"type": "json_object"}
)
print(response.choices[0].message.content)
```
**Agent tool-calling example**
```python theme={null}
tools = [{
"type": "function",
"function": {
"name": "search_internal_kb",
"description": "Search the internal knowledge base",
"parameters": {
"type": "object",
"properties": {
"query": {"type": "string"},
"top_k": {"type": "integer", "default": 5}
},
"required": ["query"]
}
}
}]
resp = client.chat.completions.create(
model="gemini-3.1-flash-lite",
messages=[{"role": "user", "content": "Find docs about the refund process."}],
tools=tools,
tool_choice="auto"
)
print(resp.choices[0].message.tool_calls)
```
### Best practices
**Production rollout tips**
1. **Smooth migration from Preview**: replace `gemini-3.1-flash-lite-preview` with `gemini-3.1-flash-lite`; shadow-traffic compare before cutover
2. **Thinking levels on demand**: keep Thinking off for routing/classification to maximize speed; turn on for multi-step reasoning
3. **Prefer structured output**: pair with `response_format={"type": "json_object"}` for robust downstream parsing
4. **Batch + Cache**: route high-volume jobs through Batch API; enable Caching on repeated context for an extra 90% off cached input
5. **Watch verbosity**: Flash Lite tends to be talkative — set `max_tokens` explicitly on cost-sensitive endpoints
## Pricing & Availability
### APIYi official-direct pricing
| Type | Price |
| -------------------------- | ------------------------------------- |
| Text / image / video input | \$0.250 per 1M tokens |
| Output | \$1.500 per 1M tokens |
| Cached input | \$0.025 per 1M tokens (\~10% of list) |
* Official direct-connection (proxy) channel, stable
* Pricing identical to Google's official rates
* Stack with Batch API for further savings
### Stackable top-up promotion (79–85％ of list)
APIYi runs ongoing top-up bonus promotions. Stacking these on top of the official-direct pricing brings effective cost for Gemini 3.1 Flash Lite down to **79–85% of list**:
| Tier | Bonus | Effective price |
| --------------- | ----- | ----------------- |
| Starter | +18% | \~**85% of list** |
| Pro | +22% | \~**82% of list** |
| High-throughput | +27% | \~**79% of list** |
See the top-up promotion page: [Recharge Promotions](/en/faq/recharge-promotions).
**Migration note**
* Google will keep `gemini-3.1-flash-lite-preview` available for some time, but **new integrations should target the GA build** `gemini-3.1-flash-lite`
* GA stabilizes the contract, but keep monitoring and fallbacks on critical paths
## Summary & Recommendations
GA Gemini 3.1 Flash Lite pushes "**speed / price / multimodality / agent capability**" to the top of its tier all at once:
* **64% faster output / 2.5× faster TTFT** vs 2.5 Flash — visibly snappier on long agent loops
* **GPQA Diamond 86.9% / MMMU Pro 76.8%** — first-tier on reasoning and multimodal at this price
* **\$0.25 / \$1.50 per 1M tokens**, plus APIYi top-up promo down to **\~79% of list**
* **Stable contract post-GA** — production-ready
**Our recommendations**
* **Teams on Preview**: cut over to the GA model name to lock in a stable contract
* **High-volume agent teams**: route routing / tool-calling / extraction nodes through Flash Lite; combine with Batch + Cache to drive unit cost to the floor
* **Multimodal lightweight teams**: cover text, image, video, audio, PDF with one model — less SDK sprawl
**Sources & data freshness**
* Google official blog: `blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-lite/`
* Google GA announcement: `cloud.google.com/blog/products/ai-machine-learning/gemini-3-1-flash-lite-is-now-generally-available`
* Model docs: `ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite`
* Benchmarks: Artificial Analysis (`artificialanalysis.ai/models/gemini-3-1-flash-lite-preview`)
* Data captured: May 9, 2026 (UTC+8)
**Get started**
Sign in to APIYi, grab an API key, point your `model` field at `gemini-3.1-flash-lite`, and you're live with a stable GA API, official-parity pricing, and top-up bonuses on top.