Gemini 2.5 Pro Preview 06 05 · gemini-2.5-pro-preview-06-05
Google · 5 Jun 2025
benchmarks
Gemini 2.5 Pro Preview 06-05 is Google's June 2025 reasoning-model preview, supporting multimodal input, long-context analysis and code generation. Stable Gemini 2.5 Pro superseded it on June 17, 2025; the specifications and historical prices here remain specific to the dated preview.
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 | 84.8%5 Jun 2025 | View |
| FrontierMath | 10.3%3 Jul 2025 | View |
| Aider PolyglotCode editing across six programming languages | 79.1%6 Jun 2025 | View |
Weighted mean of 3 verified benchmark results: GPQA Diamond 84.8%, FrontierMath 10.3%, Aider Polyglot 79.1%.
Capabilities
- Reasoning through complex mathematical, scientific and programming tasks.
- Generating, explaining and editing code across multiple programming languages.
- Analyzing text, images, audio and video inputs, with text output.
- Processing large documents and codebases within a million-token input window.
- Supporting function calling and structured output for application integrations.
What improved
- Google reported stronger coding performance than the preceding Gemini 2.5 Pro preview.
- Instruction following was improved in response to developer feedback.
- Response style and structure were refined, including more creative and better-formatted answers.
Known flaws & limitations
- Reasoning, factual answers and generated code can contain errors and require independent validation.
- Thinking adds latency and consumes billable output tokens; thinking cannot be disabled for Gemini 2.5 Pro.
- A large context window does not guarantee reliable retrieval or reasoning across every detail.
- The listed prices are historical USD rates per million tokens for prompts up to 200,000 tokens. Longer prompts cost $2.50 per million input tokens and $15 per million output tokens; output billing includes thinking tokens.
- This dated preview was superseded by stable Gemini 2.5 Pro and should not be treated as the current production model.
Risks, safety & security issues
- Untrusted documents, webpages and tool responses can introduce prompt-injection risks in agentic applications.
- Safety filtering does not eliminate the possibility of harmful or misleading outputs.
- Generated code can introduce security vulnerabilities when deployed without review and testing.
- Sensitive information supplied through application context or tools requires application-level access controls and data-handling safeguards.
Good fit for
- Historical evaluation of the June 5 preview's reasoning and coding behavior.
- Human-reviewed code generation, debugging and repository analysis.
- Long-document synthesis with source checking.
- Multimodal analysis with independently validated conclusions.
Poor fit for
- New production deployments requiring a current, supported model identifier.
- Safety-critical decisions without qualified human review.
- Latency-sensitive applications that require reasoning to be disabled.
- Autonomous execution of generated code or tool actions without permission boundaries.
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.