gemini-2.5-flash

Gemini 2.5 Flash is Google's cost-optimized model in the Gemini 2.5 generation, designed as the workhorse of the Gemini lineup — delivering strong capability at a fraction of Gemini 2.5 Pro's cost, with optional thinking mode for tasks requiring deeper reasoning. It supports a 1M-token context window, vision inputs, audio, video, tool use, and streaming. For most production workloads that don't require frontier-level capability, Gemini 2.5 Flash offers an exceptional cost-to-quality ratio and is Google's recommended default model for new applications.

Gemini 2.5 Flash is priced at $0.15 per million input tokens (non-thinking) and $0.60 per million output tokens. Thinking mode costs $3.50/1M for thinking tokens. At 1,000 input + 500 output tokens per call, it costs approximately $0.00045 per 1,000 API calls in non-thinking mode — among the lowest prices for a capable multimodal model with a 1M-token context window. The combination of low price, large context window, and optional thinking mode makes it unusually versatile: it can serve as a cheap workhorse for simple tasks and a capable reasoner for complex ones, all on the same API endpoint.

Gemini 2.5 Flash's 1M-token context window is the same as Gemini 2.5 Pro, making it well-suited to long-document workloads that would require chunking with smaller-context models. Non-thinking mode latency is approximately 200–400ms TTFT, suitable for interactive applications. Thinking mode adds 5–60 seconds depending on complexity and thinking budget. Throughput is approximately 150 tokens per second in non-thinking mode. Maximum output is 65,536 tokens. The large context window at low cost makes it particularly attractive for document processing pipelines, RAG systems, and any workload where context length is a constraint with smaller models.

Gemini 2.5 Flash is best suited to high-volume workloads requiring moderate-to-strong capability: document summarization, Q&A over long documents, code generation for common patterns, content moderation, multimodal analysis (images, video, audio), and RAG pipelines where the retrieval context is large. Thinking mode makes it suitable for more complex reasoning tasks without switching to a more expensive model. Its multimodal capability is more mature than GPT-4o mini — particularly for video and audio inputs, which GPT-4o mini does not support. For teams on Google's ecosystem (Vertex AI, Google Cloud), it integrates naturally with other Google services.

Gemini 2.5 Flash sits between Gemini 2.0 Flash (cheaper, less capable) and Gemini 2.5 Pro (more capable, more expensive) in Google's lineup. Its main competitors are other efficient-tier models from OpenAI and Anthropic, and open-weights models that can be self-hosted. For teams prioritizing cost and context window size, Gemini 2.5 Flash is the strongest option in the efficient tier. Its intelligence score of approximately 80/100 reflects strong capable-tier performance — well above typical efficient-tier models.