Google releases Gemini 3.8 Flash, its third budget model in six weeks

Gemini 3.8 Flash arrives three weeks after Gemini 3.7 Flash. Google states the new model improves coding performance by a wide margin.

Two versions launch: a general-purpose model for reasoning and coding, and a specialized cybersecurity variant named 3.8 Flash Cyber. This rapid release schedule marks the third Flash update in six weeks. Whether this pace indicates strength or serves as a distraction from the missing frontier models Gemini 3.5 Pro and Gemini 4 depends on perspective. New DeepMind head Koray Kavukcuoglu confirmed Google pursues price-performance optimisation while retaining a goal to lead on raw capability.

Google reports Gemini 3.8 Flash scores 73.7 per cent on the DeepSWE v1.1 benchmark for long-horizon software engineering tasks. This sits just below Claude Opus 5 at 74.0 per cent but ahead of Claude Sonnet 5 at 53.8 per cent, GPT-5.6 Sol at 72.7 per cent, and the previous 3.7 Flash at 65.3 per cent.

The new model also shows improvement in 3D generation. A video demonstrates a 3D game Gemini 3.8 Flash reportedly built from a single prompt using Google’s AI coding tool Antigravity. Textures were generated with Google’s Nano Banana image model.

Gemini 3.8 Flash launches at $0.75 per million input tokens and $3.75 per million output tokens, matching 3.7 Flash. Starting January 2027, prices will rise to $1.50 and $7.50 respectively. Claude Opus 5 costs $5.00 per million input tokens and $25.00 for output tokens, while GPT-5.6 Sol sits at $4.00 and $20.00. Even after introductory pricing expires, Gemini 3.8 Flash remains far cheaper per token than the top models from OpenAI and Anthropic.

Google attributes performance gains partly to extra reasoning steps on complex tasks and iterative tool calls. The model “works harder,” Google says. This increases token consumption, which partly offsets the lower per-token price. For use cases where compute efficiency matters most, Google recommends lower reasoning levels or sticking with the still-supported 3.7 Flash.

Gemini 3.8 Flash is available to developers through Google AI Studio, Google Antigravity, and Android Studio. Businesses can access it through Gemini Enterprise. Consumers find it in the Gemini app, Google Search’s AI Mode, and for paying subscribers, in Google Sheets.

Cost per task rises despite lower token rates

Independent benchmarking platform Artificial Analysis gives Gemini 3.8 Flash an Intelligence Index score of 59. This is three points above its predecessor 3.7 Flash at 56. The score puts it on par with GPT-5.6 Sol at xhigh reasoning and Grok 4.6 at medium reasoning, both also scoring 59. Artificial Analysis states the Intelligence Index gains come mainly from stronger performance on agentic benchmarks like tool use and coding tasks.

On cost per task, 3.8 Flash hits the Pareto frontier according to Artificial Analysis, coming in at $0.58 per task as the cheapest model at its intelligence level. That figure has risen about 40 per cent compared to 3.7 Flash at $0.40, even though the per-token price stayed the same. This increase likely explains why Google recommends sticking with 3.7 Flash for efficiency-focused workloads.

At high reasoning levels, 3.8 Flash produces about 300 output tokens per second with an average time per task of 2.5 minutes. That is slightly faster than GPT-5.6 Luna at 2.6 minutes and GPT-5.6 Terra at 3.3 minutes but slower than Claude Fable 5.1 at 2.1 minutes and the older 3.7 Flash at 2.2 minutes. At low reasoning levels, the time drops to about 48 seconds.

Cybersecurity model stays locked down for vetted defenders

Gemini 3.8 Flash Cyber, like its predecessor 3.5 Flash Cyber, is not publicly available. Google distributes it through the Fairwind Program to government agencies, critical infrastructure operators, and software maintainers. The model has less restrictive safety settings than the standard version because it is built for defensive cybersecurity work.

On CyberGym, the industry-standard benchmark for detecting vulnerabilities in C/C++, 3.8 Flash Cyber scores 86.2 per cent according to Google. This beats the previous 3.5 Flash Cyber at 77.5 per cent, GPT-5.6 Sol at 83.6 per cent, and GPT-5.5-Cyber at 85.6 per cent. For automated patching on the external CWE-Bench, 3.8 Flash Cyber hits 47.2 per cent Pass@1, nearly matching the leading frontier model at 47.8 per cent while costing much less.

The model also appears more resilient against prompt injection attacks. On the Gray Swan IPI benchmark, Gemini 3.8 Flash achieves an attack success rate of just 5.5 per cent. DeepSeek V4 Pro comes in at 60.1 per cent, Kimi K3 at 52.7 per cent, and Grok 4.6 at 51.8 per cent. Only Anthropic’s Claude Opus 5 does slightly better at 4.8 per cent, and Opus 5 with its additional security options scores even lower within the Claude ecosystem.

What it means

Developers gain a cheaper option for coding and 3D generation tasks that previously required more expensive frontier models. The lower per-token price helps, but the increased token usage for “harder” reasoning means total costs per task are higher than the previous Flash version. Teams prioritising speed and efficiency should stick with 3.7 Flash or use lower reasoning settings. The cybersecurity model remains restricted to vetted organisations needing defensive capabilities.