← All models

Gemini 2.5 Flash

Published Publisher Googlegemini-2.5-flash-preview-04-17

Model positioning

The 04-17 version of Gemini 2.5 Flash is an early preview model released by Google on April 17, 2025, positioned as a multimodal model that balances reasoning capability, speed, and cost. At launch, it was available through the Gemini API, Google AI Studio, and Vertex AI. (developers.googleblog.com)

Capabilities and uses

  • Supports text, image, audio, and video inputs and produces text output; the official model card lists a context window of approximately 1 million tokens and an output limit of 64K tokens. (storage.googleapis.com)
  • Supports enabling or disabling thinking and setting a thinking budget to balance response quality, latency, and cost. The launch announcement gives a budget range of 0–24,576 tokens; setting it to 0 disables thinking. (developers.googleblog.com)
  • Can be used for tasks such as document summarization, multimodal content understanding, mathematical and complex problem reasoning, and code generation. (developers.googleblog.com)

Limitations and version status

The model may hallucinate and has limitations in causal understanding, complex logical inference, and counterfactual reasoning; its adherence to the thinking budget may also be inconsistent. The model card lists a knowledge cutoff of January 2025. (storage.googleapis.com)

A subsequent Google announcement stated that the 04-17 preview endpoint was scheduled to shut down on July 15, 2025. The Vertex AI release notes also stated that the endpoint of the same name on that platform was changed to serve the 05-20 model version starting June 19, 2025. Therefore, an endpoint name does not necessarily always correspond to the original model weights; the shutdown date above is the officially announced planned date. (developers.googleblog.com)

Pricing

The comparison table in the official launch announcement lists the historical prices for the 04-17 preview in US dollars (USD), all per 1 million tokens: input $0.15; output $0.60 with thinking disabled and $3.50 with thinking enabled. These are historical preview-period prices and should not be applied to other versions or regarded as currently available prices. A subsequent official announcement stated that the preview's original prices were planned to remain in effect until the endpoint's deactivation date of July 15, 2025. Specific historical billing terms for other input modalities, caching, or batch processing have not been verified. Query date: November 18, 2026. (storage.googleapis.com)

Official sources