Gemini 3 Flash Model Card

Gemini 3 Flash is a natively multimodal reasoning model published by Google DeepMind in December 2025. It is the next iteration in the Gemini 3 series, built off Gemini 3 Pro as its foundation. The model offers configurable thinking levels to trade off quality, cost, and latency.

The model accepts text, images, audio, and video inputs with a 1M token context window and produces text output up to 64K tokens. Intended use cases include agentic workflows, everyday coding, reasoning and planning, and multimodal analysis. As of December 2025, Gemini 3 Flash "significantly outperforms Gemini 2.5 Pro across a range of benchmarks requiring enhanced reasoning and multimodal capabilities"; specific benchmark scores are hosted at deepmind.com/models/evals-methodology/gemini-3-flash rather than in this card.

Gemini 3 Flash was evaluated across reasoning, multimodal capabilities, agentic tool use, multilingual performance, and long-context benchmarks. Full methodology details and results are published externally and are not reproduced in this card. The lab notes that performance figures use improved internal evaluations and "are not directly comparable with performance results found in previous Gemini model cards."

Automated safety evaluations cover text-to-text safety (−3.1% vs. Gemini 2.5 Flash), multilingual safety (+0.1%), image-to-text safety (−2.3%), tone (+3.8%), and unjustified refusals (−10.4%). Manual red teaming was conducted by specialist teams outside the model development team; review of flagged losses found them to be "overwhelmingly either a) false positives or b) not egregious." Child safety evaluations satisfied required launch thresholds. For Frontier Safety, Google relied on Gemini 3 Pro Preview results, concluding that because Flash is less capable, it is "unlikely to reach any CCLs" outlined in their Frontier Safety Framework.

This card defers all detail on risks and mitigations to the Gemini 3 Pro model card. The lab states it "continues to improve internal evaluations, including refining automated evaluations to reduce false positives and negatives." No Flash-specific safeguard thresholds or classifier details are disclosed here.

Gemini 3 Flash was deemed acceptable for deployment "in line with the risk acceptance criteria outlined in our FSF." Distribution details mirror those of Gemini 3 Pro; this card refers readers to the Gemini 3 Pro model card for specifics. License terms, API surfaces, and access restrictions are not disclosed in this document.

Known limitations are not enumerated in this card; the document directs readers to the Gemini 3 Pro model card for that information. Acceptable usage policies are likewise deferred to the Gemini 3 Pro card.

Gemini 3 Flash supersedes Gemini 2.5 Flash and is positioned as a faster, lower-cost derivative of Gemini 3 Pro. The addition of configurable thinking levels is the primary architectural differentiator called out relative to prior Flash models. No detailed changelog is provided beyond the benchmark comparison against Gemini 2.5 Flash.