Two Variants — Flash vs Flash Cyber
Both share the same foundational intelligence but target different deployment environments.
| Aspect | Gemini 3.8 Flash | Gemini 3.8 Flash Cyber |
|---|---|---|
| Positioning | General workhorse (software engineering, agents, professional reasoning) | Cybersecurity-focused (vulnerability discovery, automated patching) |
| Access | Generally available (API, AI Studio, Antigravity, Enterprise, etc.) | Fairwind Program only — trusted defenders |
| Pricing | $0.75/$3.75 (intro) → $1.50/$7.50 | Not public (Fairwind Program) |
| Use | Long-horizon coding, autonomous agents, multi-step reasoning | Autonomous vulnerability discovery and automated patching (defensive) |
| Safety | Standard safeguards (CBRN + cyber offense mitigations) | More permissive cyber mitigations for trusted defenders |
Source: Google blog (Introducing Gemini 3.8 Flash and 3.8 Flash Cyber) (verified 2026-09-04)
Pricing (intro vs. standard)
Verified against Google's official pricing page (ai.google.dev). All prices in USD per 1M tokens. Output pricing includes thinking tokens.
| Tier | Input | Output | Cached Input | Notes |
|---|---|---|---|---|
| Standard (intro, thru 12/31) | $0.75 | $3.75 | $0.075 | Through Dec 31, 2026 |
| Standard (from 1/1) | $1.50 | $7.50 | $0.15 | From Jan 1, 2027 |
| Batch / Flex (intro, thru 12/31) | $0.375 | $1.875 | $0.0375 | Half of Standard |
| Batch / Flex (from 1/1) | $0.75 | $3.75 | $0.075 | From Jan 1, 2027. Half of Standard |
| Priority (intro, thru 12/31) | $1.35 | $6.75 | $0.135 | High priority. 1.8× Standard |
| Priority (from 1/1) | $2.70 | $13.50 | $0.27 | From Jan 1, 2027. 1.8× Standard |
Primary sources: ai.google.dev/gemini-api/docs/pricing · Google Blog (verified 2026-09-04)
Model Specifications (Gemini 3.8 Flash)
Sources: Google AI Gemini 3.8 Flash Model Docs (verified 2026-09-04). Knowledge cutoff not stated in official docs. Supports caching, code execution, file search, search/maps grounding, structured outputs and URL context.
Performance — Big Gains Over 3.7 Flash, Approaching Frontier Models
3.8 Flash "works harder" — on complex tasks it runs extra reasoning steps and iterative tool calls, which can increase token use at higher effort levels.
- DeepSWE v1.1: 73.8% — long-horizon software engineering, outperforming most larger frontier models (per OpenAI's comparison table).
- HLE-Verified: 54.9% — multi-step reasoning across STEM, humanities and professional fields (up from 3.7 Flash).
- GPQA Diamond: 95.3% — graduate-level science reasoning (beats 3.7 Flash).
- Vals Finance Agent V2 / Harvey Legal Agent Benchmark: outperforms 3.7 Flash and other frontier models in specialized domains.
- Artificial Analysis Intelligence Index: 58.7 — up from 3.7 Flash (56).
- FrontierCode 1.1 Main: 43.6% / Extended: 56.3% (per OpenAI's comparison table).
Google positions 3.8 Flash as "our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows." For compute-efficiency-first workloads, use lower effort levels or stay on Gemini 3.7 Flash, which remains fully supported.
Sources: Google Blog (evals) · OpenAI GPT-6 Astra announcement (comparison table) (verified 2026-09-04)
3.8 Flash Cyber — Cybersecurity Performance
Purpose-built for defenders, prioritizing vulnerability fixing over offensive capabilities like exploitation.
- CyberGym (industry standard for vulnerability discovery): frontier-level autonomous discovery, surpassing 3.5 Flash Cyber and larger frontier models.
- Internal benchmark (20 languages, complex codebases): success rate above 70%.
- CWE-Bench pass@1: 47.2% — on the Pareto frontier (leading frontier model 47.8%) at significantly lower cost.
- Real-world: Chrome Security team — 2.6× more correct patches vs larger commercial models. Wiz — +7.5–9.7% recall at 2.3–5.2× lower cost. Google Cloud Vulnerability Research found a critical vulnerability in under 2 hours (usually months).
Sources: Google Blog · Fairwind Program (verified 2026-09-04)
Competitor Price Comparison
The intro price of $0.75/$3.75 is among the cheapest in the Flash tier. All prices in USD per 1M tokens (Standard, text input).
| Model | Provider | Input | Output | vs 3.8 Flash | Notes |
|---|---|---|---|---|---|
| Gemini 3.8 Flash | $0.75→$1.50 | $3.75→$7.50 | — | Intro/standard. Multimodal | |
| Gemini 3.7 Flash | $0.75→$1.50 | $3.75→$7.50 | Same | Still supported for efficiency-first work | |
| Gemini 3.6 Flash | $0.75→$1.50 | $3.75→$7.50 | Same | Same intro price applies | |
| Gemini 3.5 Flash | $1.50 | $9.00 | 3.8 cheaper | 50% cheaper input, 58% cheaper output (intro) | |
| GPT-5.6 Luna | OpenAI | $0.20 | $1.20 | Luna cheaper | Text + image only. Cheapest in absolute terms |
| GPT-6 Astra | OpenAI | $10.00 | $50.00 | 3.8 cheaper | Flagship tier (different class) |
| Grok 4.6 | xAI | $2.00 | $6.00 | 3.8 cheaper | 62% cheaper input, 37% cheaper output (intro) |
* Gemini 3.8 Flash's intro price ($0.75/$3.75) runs through Dec 31, 2026; standard $1.50/$7.50 follows. GPT-5.6 Luna ($0.20/$1.20) is cheaper in absolute terms but serves high-throughput niches; 3.8 Flash targets high-difficulty coding, agents and professional reasoning.
Sources: official pricing pages from each provider (verified 2026-09-04). See full pricing comparison. Gemini 3.7 Flash details →
Ideal Use Cases
Coding, agents and specialized multi-step reasoning.
Long-horizon Coding
DeepSWE 73.8%. Autonomously solves complex engineering problems end to end.
Autonomous Agents
Multi-step planning and iterative tool calls. Built for long-running agentic loops.
Finance & Legal Reasoning
Strong Vals Finance Agent V2 / Harvey LAB. Expert-domain analysis and reporting.
Multi-step Reasoning
HLE-Verified 54.9%. STEM, humanities and professional fields.
Web Dev & UI Generation
Builds games and apps from a single prompt in Google Antigravity.
Long & Multimodal Processing
1M context, PDF/video/audio input for large-document Q&A.
Gemini 3.x Family Selection Guide
| Model | Input | Output | When to Use |
|---|---|---|---|
| Gemini 3.1 Pro Preview | $2.00 | $12.00 | Hardest reasoning, agents, complex multi-step problems |
| Gemini 3.8 Flash ⬅ | $0.75→$1.50 | $3.75→$7.50 | Coding, autonomous agents, professional reasoning (half price thru 2026) |
| Gemini 3.7 Flash | $0.75→$1.50 | $3.75→$7.50 | Efficiency-first coding/agents (still supported) |
| Gemini 3.5 Flash | $1.50 | $9.00 | Standard Flash tier. Slightly higher output than 3.8 |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | High-throughput, cost-sensitive, multimodal bulk processing |
💡 Strategy: use Flash-Lite for routine high-frequency tasks, 3.8 Flash for production coding/agents, and Pro for the hardest problems. 3.8/3.7/3.6 are available at $0.75/$3.75 through the end of 2026.
Estimate Your Actual Costs
Compare token costs for Gemini 3.8 Flash and all major models. Just enter your token counts.
🧮 Open Token Calculator →📊 Full Pricing Comparison · ⚡ Compare with Gemini 3.7 Flash · 🗓️ Release Timeline · 🔗 Google Official Pricing
⚠️ Disclaimer
- Pricing verified against Google's official AI page (ai.google.dev) on 2026-09-04. The intro price runs through Dec 31, 2026, then $1.50/$7.50. Always check official pages before contracting.
- Gemini 3.8 Flash Cyber is available only through the Fairwind Program (trusted defenders); it is not on the public API and has no public pricing.
- This site is informational only. Not affiliated with any provider.
- Performance claims are based on Google's published data. Your mileage may vary.