- Instruction Following
Gemini 3.5 Flash is Google’s natively multimodal reasoning model optimized for very low latency and cost while maintaining frontier‑level performance, particularly for coding and agentic workflows.
Powered by Google
Gemini 2.5 Flash Lite Preview 09-2025
- Instruction Following
Gemini 2.5 Flash Lite Preview 09-2025 is a lightweight preview variant of Google’s Gemini 2.5 Flash-Lite model, optimized for fast, cost-efficient multimodal inference with long-context support. It offers text outputs from text, image, video, audio, and PDF inputs while showcasing improvements over the earlier Flash-Lite release.
Providers Comparison
Speed
Output tokens per second · Higher is better
L
LLM API
S
SiliconFlow (FP8)
G
GMI
C
CoreWeave
N
Novita
D
DeepSeek
P
Parasail (FP8)
D
DeepInfra (FP4)
End-to-End Response Time
Seconds · Lower is better
L
LLM API
S
SiliconFlow (FP8)
G
GMI
C
CoreWeave
N
Novita
D
DeepSeek
P
Parasail (FP8)
D
DeepInfra (FP4)
Price
USD per 1M tokens (blended) · Lower is better
L
LLM API
S
SiliconFlow (FP8)
G
GMI
C
CoreWeave
N
Novita
D
DeepSeek
P
Parasail (FP8)
D
DeepInfra (FP4)
About the model
What is Gemini 2.5 Flash Lite Preview 09-2025?
Gemini 2.5 Flash Lite Preview 09-2025 is a Google Gemini API model variant that provides a preview of updated Flash-Lite capabilities as of September 2025. It is mainly used for low-latency, high-throughput applications such as chatbots, agents, and tools that need long-context reasoning over large text or multimodal documents. It also targets developer workloads like batch processing, retrieval-augmented generation, and structured outputs using function calling and file search. It belongs to the Gemini 2.5 Flash-Lite family and is offered alongside the stable gemini-2.5-flash-lite model as a preview version.
Model capabilities
5 Core Capabilities
-
Multimodal Input
Accepts very long-context inputs across text, code, images, audio, and video while generating coherent text-only responses efficiently.
-
Conversational Chat
Handles interactive dialogue, following instructions and maintaining context over extended conversations with low latency and low cost.
-
Grounded Reasoning
Enhances answers using grounding with Google Search, improving factuality and up-to-date knowledge in supported use cases.
-
Global Language Support
Supports many input and output languages, enabling multilingual applications for users across diverse regions and locales.
-
Image Understanding
Analyzes images within multimodal prompts to extract visual details, interpret content, and incorporate findings into generated text.
Use cases
6 Most Valuable Use Cases
- High-volume Chatbots
- Streaming Data Summaries
- Search Query Expansion
- Alert Log Monitoring
- E-commerce Product Support
- Lightweight On-device Inference
Transparent pricing
Cost Comparison
Up to ~60% cheaper and faster than comparable Gemini-class LLMs
| Provider | Region | Latency | Throughput | Uptime | Input ($/1M) | Output ($/1M) | Context |
|---|---|---|---|---|---|---|---|
| LLM API BEST | Global | 120ms | 120 tps | 99.99% | $0.05 | $0.10 | 256K |
| Global | ~220ms | ~60 tps | ~99.9% | ~$0.12 | ~$0.24 | ~256K | |
| OpenRouter | Global | ~260ms | ~45 tps | ~99.9% | ~$0.14 | ~$0.28 | ~128K |
| Together AI | US East | ~250ms | ~50 tps | ~99.9% | ~$0.13 | ~$0.26 | ~128K |
| Fireworks AI | US West | ~240ms | ~55 tps | ~99.9% | ~$0.13 | ~$0.25 | ~128K |
Performance benchmarks
Technical Specifications
| Metric | Gemini 2.5 Flash Lite Preview 09-2025 | GPT-4.1 Mini (OpenAI) | Claude 3.7 Haiku (Anthropic) |
|---|---|---|---|
| Avg Latency | ~180ms | ~220ms | ~250ms |
| Context Window | 128K | 128K | 200K |
| Input Price ($/1M) | $0.03 | $0.15 | $0.25 |
| Output Price ($/1M) | $0.06 | $0.60 | $0.75 |
| Max Output Tokens | 8K | 4K | 8K |
| Throughput | ~500 tps | ~200 tps | ~180 tps |
| Uptime | 99.9% | 99.9% | 99.9% |
30-day usage via LLM API
- 6.8B
- Prompt tokens processed (last 30 days)
- 420M
- Completion tokens generated (last 30 days)
- 19.5M
- API requests served (last 30 days)
- 99.96%
- Average uptime over the last 30 days
Architecture & Integration
Why Build on LLM.API?
One unified API. Every major model. Built-in reliability, cost control, and observability.
-
Unified AI Routing
Automatically route each request to the optimal model across providers based on latency, cost, and capability—without changing your integration or redeploying code.
One endpoint, every model. -
Cost-Aware Orchestration
Set cost ceilings and policies once, then let LLM.API select the cheapest model that still meets your quality and latency requirements in real time.
Optimize spend by default. -
Resilient Fallbacks
Define multi-provider fallback chains so when a model or region fails, traffic seamlessly fails over—no downtime, no emergency redeploys.
Stay online, automatically. -
Deep Observability
Get per-request traces, latencies, errors, and token usage across all providers in one place, with structured logs ready for your existing monitoring stack.
See every token, everywhere. -
Task-Level Abstractions
Express work as high-level tasks—chat, extraction, tools, agents—while LLM.API handles prompts, models, and retries so your code stays clean and consistent.
Code to tasks, not models. -
High-Throughput Batching
Submit large batches in a single call and let LLM.API handle parallelization, rate limits, retries, and aggregation for massive throughput and lower unit costs.
Scale runs, not complexity.
Decision guide
When to Use — When NOT to Use
Use it if...
- You need a very low-cost model for high-volume requests and experimentation.
- You need fast responses for lightweight chatbots, assistants, or simple interactive tools.
- Your use case involves basic text generation, summarization, or rewriting with modest complexity.
- Your use case involves simple multi-turn conversations without heavy long-term memory requirements.
- You need a small, responsive model to prototype features before upgrading to stronger variants.
- Your use case involves low-stakes tasks where minor reasoning errors are tolerable.
Avoid if...
- You need state-of-the-art reasoning quality for complex analysis, planning, or problem solving.
- Your workload requires highly reliable code generation, debugging, or large-codebase understanding.
- You need advanced tool orchestration, multi-step agents, or robust function-calling workflows.
- Your workload requires strong tool-using agents handling intricate, multi-step decision workflows.
- You need maximum answer quality and nuance for customer-facing, high-stakes user interactions.
- Your workload requires strict, extensively evaluated safety and controllability guarantees at scale.
FAQ
Frequently Asked Questions
-
What is Gemini 2.5 Flash Lite Preview 09-2025?
Gemini 2.5 Flash Lite Preview 09-2025 is a Google Gemini model variant optimized for low-latency, cost-efficient multimodal generation in public preview.
-
What is the context window of Gemini 2.5 Flash Lite Preview 09-2025?
Gemini 2.5 Flash Lite Preview 09-2025 supports up to 1,048,576 input tokens and 65,535 output tokens, giving it roughly a 1M token context window.
-
What modalities does Gemini 2.5 Flash Lite Preview 09-2025 support?
Gemini 2.5 Flash Lite Preview 09-2025 accepts text, code, images, audio, and video as input and generates text-only outputs.
-
What is Gemini 2.5 Flash Lite Preview 09-2025 best suited for?
It is best for high-throughput, latency-sensitive applications like chatbots, agents and lightweight multimodal understanding where low cost and speed matter more than peak quality.
-
How fast is Gemini 2.5 Flash Lite Preview 09-2025 compared to other Gemini 2.5 models?
Flash Lite Preview is tuned for lower latency and higher throughput than Gemini 2.5 Pro, at slightly lower raw reasoning and generation quality.
-
How is Gemini 2.5 Flash Lite Preview 09-2025 priced?
On Google Cloud it uses pay-as-you-go token-based billing with discounted input tokens when context caching is used; LLM.API applies its own unified pricing.
-
How do I access Gemini 2.5 Flash Lite Preview 09-2025 via LLM.API?
Call the LLM.API chat or completion endpoint with the provider set to Google and the model set to "gemini-2.5-flash-lite-preview-09-2025".
-
How does Gemini 2.5 Flash Lite Preview 09-2025 compare to Gemini 2.5 Flash Lite GA?
The preview model shares the same core architecture but has an earlier lifecycle, fewer supported features, and a scheduled discontinuation date of July 9, 2026.
-
What are the main limitations of Gemini 2.5 Flash Lite Preview 09-2025?
It does not support Gemini Live API, supervised fine-tuning, or chat-completions endpoints and is constrained by a January 2025 knowledge cutoff.
-
Can I use Gemini 2.5 Flash Lite Preview 09-2025 for real-time voice streaming?
No, this preview model is not exposed through Gemini Live API, so it cannot be used for real-time streaming audio conversations.
Compare
Competitive Models
- Instruction Following
Qwen3 Max is Qwen’s flagship trillion-parameter large language model, offered as a high-end proprietary API model. It is designed to deliver state-of-the-art performance across reasoning, coding,…
- Instruction Following
Qwen3 VL 235B A22B Instruct is a 235B-parameter Mixture-of-Experts vision-language model from Qwen, offering open-weight, long-context (≈256K) multimodal reasoning over text, images, and video. It is…