Gemini 2.5 Pro Preview 05 06 · gemini-2.5-pro-preview-05-06
Google · 6 May 2025
benchmarks
Gemini 2.5 Pro Preview 05 06 is a dated reasoning-model preview emphasizing coding, frontend development, and multimodal understanding. It was superseded by later Gemini 2.5 Pro releases, including the stable release announced on June 17, 2025, and should not be confused with the current production model.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 66.7%3 Jun 2025 | View |
| MATH Level 5 | 95.9%8 May 2025 | View |
| Aider PolyglotCode editing across six programming languages | 76.9%7 May 2025 | View |
Weighted mean of 3 verified benchmark results: GPQA Diamond 66.7%, MATH Level 5 95.9%, Aider Polyglot 76.9%.
Capabilities
- Reasoning across complex coding, mathematics, and analytical tasks.
- Frontend development and generation of interactive web applications.
- Code transformation, editing, and debugging.
- Understanding text, images, audio, and video as inputs, with text output.
- Long-context analysis of documents and codebases.
- Function calling for integration with application tools.
What improved
- Improved frontend and interactive web application development over the preceding Gemini 2.5 Pro preview.
- Improved code transformation and editing.
- Improved instruction following in coding workflows.
- Improved function calling for coding applications.
Known flaws & limitations
- This dated preview was superseded and is not the stable Gemini 2.5 Pro release.
- Native output is text rather than generated images or audio.
- Historical paid API rates increased above 200,000 input tokens to $2.50 per million input tokens and $15 per million output tokens.
- Historical output-token billing included thinking tokens, increasing costs for reasoning-intensive tasks.
Risks, safety & security issues
- Google documented threat actors repurposing earlier Gemini services for phishing and reconnaissance; that reporting does not identify this preview.
- Untrusted documents, retrieved pages, and tool responses can contain prompt-injection instructions.
- Generated factual claims, citations, and code require independent validation before consequential use.
- Tool-connected applications require external authorization and least-privilege controls to limit harmful actions.
Good fit for
- Historical evaluation of May 2025 coding and reasoning performance.
- Frontend prototyping and interactive application development in historical comparisons.
- Comparative analysis of long-context document and code understanding.
- Research into changes between Gemini 2.5 Pro preview releases.
Poor fit for
- New production deployments requiring a currently supported model identifier.
- Native image or audio generation.
- Security-sensitive autonomous actions without external authorization controls.
- Publication of factual or technical outputs without independent review.
References
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.