Gemini 3.8 Flash

// Overview

Gemini 3.8 Flash is the Flash-tier Gemini model Google released on 2 September 2026 as the API model `gemini-3.8-flash`. Google describes it as "our best reasoning & coding model yet, at the same speed and low cost of 3.7", and points it at long-horizon software engineering, autonomous agents, and enterprise workflows that need factual rigor. It accepts text, image, audio and video input across a 1 million-token context window and returns up to 64K output tokens.

Google's launch post says Gemini 3.8 Flash outperforms most larger frontier models on DeepSWE v1.1 for solving complex engineering problems end to end, and that it beats Gemini 3.7 Flash and other frontier models on the Vals Finance Agent V2 and Harvey legal agent benchmarks. On HLE-Verified it scores 54.9%, which Google offers as evidence of multi-step reasoning across STEM, humanities and professional fields. Google also reports a significant improvement in prompt-injection robustness as measured by Gray Swan.

Reasoning depth is selectable through three thinking levels — low for latency-critical work, medium as the default for complex code and agentic use, and high to maximise reasoning on deep mathematical and multi-step problems. The model is served through Google AI Studio, Android Studio and Google Antigravity for developers, Gemini Enterprise for companies, and the Gemini app and Google Search for Google AI Pro and Ultra subscribers. Its knowledge cutoff is March 2026, with some domains limited to January 2025.

Google launched a second model alongside it: Gemini 3.8 Flash Cyber, a security variant Google calls its most capable cybersecurity model. Google reports it surpasses 3.5 Flash Cyber and significantly larger frontier models on CyberGym for autonomous vulnerability discovery, exceeds a 70% success rate finding real vulnerabilities across codebases in 20 programming languages, and reaches 47.2% pass@1 on CWE-Bench for patching against a leading frontier model's 47.8% at much lower cost. The Cyber model is not generally available — it ships only to vetted defenders through Google's new Fairwind Program.

Pricing is $0.75 per million input tokens and $3.75 per million output tokens through 31 December 2026, rising to $1.50 / $7.50 on 1 January 2027. That matches the introductory rate Google set for Gemini 3.7 Flash. A free tier is available in Google AI Studio.

// Benchmarks

1. HLE-Verified (multidisciplinary expert reasoning)54.9%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Introductory rate through 31 December 2026. From 1 January 2027: $1.50 input, $7.50 output. A free tier is available in Google AI Studio.

Pricing source ↗

// Strengths

• 54.9% on HLE-Verified for multi-step reasoning across STEM, humanities and professional fields
• Outperforms most larger frontier models on DeepSWE v1.1 for end-to-end engineering problems, per Google
• Beats Gemini 3.7 Flash on the Vals Finance Agent V2 and Harvey legal agent benchmarks, per Google
• 1M-token multimodal context (text, image, audio, video) with 64K output tokens
• Selectable thinking levels (low / medium / high) to trade latency against reasoning depth
• Improved prompt-injection robustness as measured by Gray Swan
• $0.75 / $3.75 per million tokens through 2026 — the same introductory rate as Gemini 3.7 Flash

// Best for

• Reach for it for long-horizon software engineering that has to run to completion without supervision
• Reach for it for autonomous agents that plan over many steps and orchestrate several tools
• Reach for it for enterprise workflows where accuracy and factual rigor matter more than raw speed
• Reach for it for repo-scale and long-document work that needs the full 1M-token window

// How to access

// Gemini Flash — every version

The full lineage of the Gemini Flash line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.