← All models

Gemini 2.5 Flash Preview

Published Publisher Googlegemini-2.5-flash-preview-05-20

Model positioning

Gemini 2.5 Flash Preview is a preview model released by Google on May 20, 2025, positioned as a high-efficiency model balancing speed, cost, and reasoning capabilities. Compared with previous versions, Google reported improvements on reasoning, multimodal, code, and long-context benchmarks. (blog.google)

Capabilities and uses

  • Reasoning and control: Supports thinking budgets, allowing developers to trade off quality, latency, and cost, or turn off thinking; the Gemini API and Vertex AI provide thinking summaries. (blog.google)
  • Code and complex tasks: Suitable for code generation, programming assistance, and applications involving complex reasoning. (blog.google)
  • Multimodal and long context: The official announcement reported improvements in multimodal and long-context capabilities. Google also released separate previews of native audio dialogue and text-to-speech around the same time; these features should not be directly equated with the full capabilities of this model ID. (blog.google)

Release and limitations

  • Release date: May 20, 2025. At launch, it was available in preview through Google AI Studio and Vertex AI. (blog.google)
  • Version status: On June 17, 2025, Google stated that the generally available version of Gemini 2.5 Flash was unchanged from the 05-20 preview model, but used a different model endpoint and pricing. (developers.googleblog.com)
  • Endpoint shutdown: The official Gemini API deprecation table lists this preview model as shut down, with a date of November 18, 2025; the Vertex AI release notes list plans to shut down that platform's preview endpoint after July 15, 2025. These are different platforms. (aistudio.google.com)
  • Usage limitations: The preview is not suitable as a production endpoint relied on for stable, long-term availability. Model outputs may still contain errors, and users should verify important results.

Pricing

Date checked: November 18, 2026. This preview endpoint has been shut down, and currently applicable input and output prices have not been verified.

In its June 2025 pricing update for the generally available version, Google noted that the previous preview pricing included rates of USD 0.15 per million input tokens and USD 3.50 per million output tokens, and distinguished between thinking and non-thinking prices; the announcement did not fully list the non-thinking output rate or the conditions for each input modality, so those details have not been verified. These are historical prices, not currently available purchase quotes; pricing for the generally available version cannot be directly applied to this preview endpoint. (developers.googleblog.com)

Official sources