Gemini CLI users rejoice: Gemini 3 Flash Preview is now available. Let’s take a look at the numbers!

What we measured:

• Sigmascore — the overall measure of an agent’s real-world coding performance
• Accuracy — how often outputs meet quality thresholds
• Consistency — how often outputs remain useful even when not fully completing a task
• Speed — how quickly tasks are completed

Each score is assigned a tier based on how close they are with respect to the margin of error. The top-scoring group of agents are in Tier 1, the next-best group are in Tier 2, and so on.

Gemini CLI (Gemini 3 Flash Preview) ranks #2 on the Sigmabench benchmark:

• It outperforms both Gemini 2.5 Flash and Gemini 3 Pro Preview on every Sigmabench metric
• It’s Tier 2 on Accuracy, statistically tied with the likes of Claude Code (Opus 4.5)
• It’s Tier 1 in Consistency, which only the Codex family has achieved thus far
• It’s Tier 2 in Speed, second only to Cursor CLI (Composer-1)

Comparisons

Gemini CLI: Gemini 3 Flash Preview vs Gemini 2.5 Flash

For Gemini CLI users, Gemini 3 Flash Preview is a huge improvement over Gemini 2.5 Flash. It’s still just as fast, but it’s a lot more accurate and consistent.

Gemini CLI: Gemini 3 Flash Preview vs Gemini 3 Pro Preview

Surprisingly, Gemini 3 Flash Preview outperforms Gemini 3 Pro Preview on every Sigmabench metric. Note that both models were tested on the same Gemini CLI version (0.21.1), released on 18 December 2025. Also note that we observed a significantly higher number timeouts (77 for Pro vs 43 for Flash) when running Gemini 3 Pro Preview which hurts both its Accuracy and Consistency scores.

It’s also woth mentioning that our benchmarking of Gemini CLI has run into a known issue with loop detection. We will be updating our benchmark results once this issue is resolved.

Key Insights

• Gemini 3 Flash Preview is a big step forward for Gemini CLI users. Its Tier 1 Consistency places it alongside the Codex family, while maintaining strong Accuracy and competitive Speed, resulting in a #2 overall Sigmascore on Sigmabench.

• Gemini 3 Flash Preview is a clear upgrade over prior Gemini models. It significantly improves Accuracy and Consistency over Gemini 2.5 Flash, and even outperforms Gemini 3 Pro Preview across all metrics, despite Pro’s positioning as a higher-tier model.

• Lower timeout rates and more dependable partial outputs give Gemini CLI (Gemini 3 Flash Preview) a practical edge, making it one of the most dependable choices for iterative, CLI-based coding workflows today.