AI Radar

Gemini 2.5 Pro

Available · Language, Multimodal, Reasoning

Gemini 2.5 Pro is a multimodal reasoning model in Google's Gemini Pro line, first released as an experimental model on 25 March 2025 and generally available from 17 June 2025 in the Gemini API and the Gemini app. Google describes it as a thinking model that reasons through its thoughts before responding; since 18 September 2026 API access is limited to earlier users. [1] [2] [3]March 25, 2025; June 17, 2025; September 18, 2026 Primary source

Timeline of Gemini 2.5 Pro →

Claims and evidence

  • Released Released as an experimental model (gemini-2.5-pro-exp-03-25); generally available on 2025-06-17. Primary source[1] [3]March 25, 2025
  • Paid public preview (gemini-2.5-pro-preview-03-25) Other sources give: [4]Gemini 2.5 Pro models: gemini-2.5-pro-preview-03-25 Release date given on the deprecations page; it is earlier than the experimental release of 2025-03-25 reported by the release notes and the announcement. Primary source[3]April 4, 2025
  • Updated preview (gemini-2.5-pro-preview-05-06) Primary source[3]May 6, 2025 [4]Gemini 2.5 Pro models
  • Updated preview (gemini-2.5-pro-preview-06-05) Primary source[3]June 05, 2025 [4]Gemini 2.5 Pro models
  • Generally available (gemini-2.5-pro) Primary source[3]June 17, 2025 [2] [5] [4]Gemini 2.5 Pro models [6]Models available for at least 12 months after release
  • Status AvailableNot deprecated in the Gemini API (no shutdown date announced), but since 2026-09-18 access to the 2.5 models is limited to users who used them before. On Gemini Enterprise Agent Platform the retirement of gemini-2.5-pro is scheduled for 2026-10-20, with gemini-3.8-flash or gemini-3.5-flash as replacements. Primary source[7]notice at the top of the page [4]Gemini 2.5 models [3]September 18, 2026 [6]Models available for at least 12 months after release
  • Successor of Gemini 2.0 Pro EditorialEditorial link along the Pro tier: the announcement names Gemini 2.0 Flash Thinking as Google's first thinking model and compares coding with Gemini 2.0 in general, but names no Pro predecessor; Gemini 2.0 Pro (experimental) was the most recent Pro model. Primary source[1]
  • Change · Reasoning Built as a thinking model that reasons through its thoughts before responding; Google says it will build thinking into all its models.Compared with Gemini 2.0 Pro Primary source[1]
  • Change · Other Google reports a large step up in coding performance over Gemini 2.0.Compared with Gemini 2.0 Pro Primary source[1]
  • Change · Context length Launched with a 1 million token context window and 2 million announced as coming soon; Gemini 2.0 Pro had 2 million.Compared with Gemini 2.0 Pro Primary source[1] [8]2.0 Pro Experimental section
  • Input text, image, audio, videoThe Gemini API also accepts PDF files. Primary source[7]Supported data types [9]Model Information: Inputs
  • Output text Primary source[7]Supported data types [9]Model Information: Outputs
  • Feature Reasoning modeThinking model; the stable version has adaptive thinking with a configurable thinking budget. Primary source[1]second paragraph [7]Capabilities: Thinking
  • Feature Function calling Primary source[7]Capabilities
  • Feature Tool useCode execution, search grounding and URL context in the Gemini API. Primary source[7]Capabilities
  • Context window 1,048,576 tokensThe launch post gave 1 million tokens, with 2 million announced as coming soon. Primary source[7]Token limits: input token limit
  • Parameters Not disclosedNot disclosed by Google. No claim madeNo source
  • Access API, consumer appGoogle AI Studio and Vertex AI (Gemini API) and the Gemini app. Primary source[1]Introducing Gemini 2.5 Pro [2]

Lineage

Predecessors

Successors

Based on

Not derived from another model.

Variants and derived

None recorded.

Siblings

None recorded.

All ancestors

All descendants

Variants

No variants recorded in this record.

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from Google on AI Radar →