Gemini 3.8 Flash Launches: Best Reasoning Flash Yet

Google DeepMind released Gemini 3.8 Flash on September 2 — the third Flash-model release in just six weeks — keeping the same introductory pricing as 3.7 Flash ($0.75/$3.75 per million tokens) while claiming its best reasoning and coding Flash yet. A companion model, Gemini 3.8 Flash Cyber, targets cybersecurity defenders through the new Fairwind Program.

Key facts

• Release date: September 2, 2026 (GA on Gemini API and Google AI Studio same day).
• Pricing: Introductory $0.75 per million input tokens / $3.75 per million output tokens — identical to 3.7 Flash’s intro rates. Google says this is “the same introductory price as 3.7 Flash.”
• Two variants:
  • Gemini 3.8 Flash — “our most intelligent workhorse model yet for coding and agents,” engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows.
  • Gemini 3.8 Flash Cyber — a specialized cybersecurity model with “frontier-level performance in vulnerability detection and automated patching,” available to trusted defenders through the new Fairwind Program.
• Context window: 1,048,576 input tokens; 65,536 output tokens.
• Knowledge cutoff: March 2026.
• Capabilities: Text, image, video, audio, PDF inputs; code execution, computer use (preview), file search, function calling, grounding, structured outputs, thinking (low/medium/high), caching, search grounding, URL context, structured outputs.
• Fairwind Program: A new proactive cyber defense initiative for governments and enterprises; 3.8 Flash Cyber available to “trusted defenders” — not broadly released.

Benchmark claims (per Google)

Google claims significant improvements over 3.7 Flash:

• Software engineering: Completes more than 3x as many tasks in long-running, document-heavy workflows vs 3.7 Flash.
• Agentic benchmarks: Outperforms 3.7 Flash and “other frontier models” on Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark.
• HLE-Verified: 54.9% (demonstrating multi-step STEM/humanities/professional reasoning).
• Safety: Overall similar to 3.7 Flash; multilingual safety regressed slightly (+5.4pp); unjustified refusals improved (+1.1pp).
• CyberGym: 3.8 Flash Cyber achieves “frontier-level performance in autonomous vulnerability discovery,” surpassing previous Cyber model and “significantly larger frontier systems.”

Pricing context

The introductory $0.75/$3.75 rate runs through at least December 31, 2026 (same window as 3.7 Flash). After that, pricing reverts to $1.50/$7.50 — double the intro rate. This is the third Flash release in six weeks (3.6 Flash Jul 21, 3.7 Flash Aug 13, 3.8 Flash Sep 2), and Google has held the intro price steady across all three.

What changed vs 3.7 Flash

Google emphasizes three vectors of improvement:

1. Long-horizon reasoning: “Sustained reasoning capabilities” for multi-step agentic tasks — completing 3x more long-running workflows.
2. Agentic execution: “Navigates roadblocks and resolves coding and real-world issues with accuracy” — optimized for autonomous agents.
3. Cyber specialization: 3.8 Flash Cyber targets vulnerability discovery and automated patching via the Fairwind Program (restricted to trusted defenders, not public).

What to watch

• Whether the 3x workflow completion claim holds in independent evals (DeepMind cites Glean and Loopit as early partners).
• Whether Fairwind Program adoption scales — this is Google’s answer to Anthropic’s Defender Advantage Fund and OpenAI’s Daybreak cyber program.
• Whether the 6-week release cadence continues (3.6 Flash Jul 21 → 3.7 Flash Aug 13 → 3.8 Flash Sep 2).

Official source

Google DeepMind Blog — Introducing Gemini 3.8 Flash and 3.8 Flash Cyber (September 2, 2026); Gemini 3.8 Flash Model Card (Published September 2, 2026); Gemini API Release Notes (September 2, 2026).