Google's Gemini 3.8 Flash Thinks Harder, But That Extra Effort Will Cost You More
Same price per token as its predecessor, but Google warns the new model uses more tokens to get better results. Here is what developers and everyday subscribers need to know.
Key points
- Google launched Gemini 3.8 Flash in May 2025, just weeks after Gemini 3.7 Flash.
- Per-token pricing stays the same: $0.75 per million tokens fed in, $3.75 per million tokens generated.
- Early testing found total task costs running roughly 40% higher than 3.7 Flash because the model produces more output.
- The model topped two industry benchmarks for software coding and legal AI agents.
- A restricted version called Gemini 3.8 Flash Cyber is available only to governments and vetted security partners.
Google's newest AI model is faster and smarter than its predecessor. It'll also likely run up a bigger bill.
Gemini 3.8 Flash arrived just weeks after Gemini 3.7 Flash, and Google says the upgrade is straightforward: the model "works harder." On complex tasks it carries out more reasoning steps and calls outside tools repeatedly until it finds a good answer. Think of it like the difference between an employee who hands you a first draft and one who reads it back, catches the gaps, then sends revised versions before calling it done.
What does it actually cost?
The sticker price is unchanged. Google charges $0.75 per million input tokens and $3.75 per million output tokens. A token is roughly three-quarters of a word.
But Google flags the catch itself: "the model might use more tokens to maximize performance, especially at higher effort levels." Early numbers from AI benchmarking firm Artificial Analysis, first highlighted by The Verge AI, put real-world costs about 40% above 3.7 Flash. The firm traced that to a 30% rise in words generated per task, plus extra back-and-forth on agentic work. An AI agent, here, means software that handles multi-step jobs on its own, like filing a support ticket or running a web search and digesting the result.
Developers who want to hold the line on spending can stay on Gemini 3.7 Flash. Google confirmed both models remain available. We've been tracking AI pricing shifts since July, and this pattern of flat rates masking higher real-world spend is one we've flagged repeatedly in our 18 AI pricing stories this quarter.
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Est. Cost vs 3.7 Flash |
|---|---|---|---|
| Gemini 3.7 Flash | $0.75 | $3.75 | baseline |
| Gemini 3.8 Flash | $0.75 | $3.75 | ~40% higher in practice |
Is it actually better at real work?
Yes, on the tasks Google chose to test. The model beat competitors on three external benchmarks: DeepSWE v1.1 for software engineering, Vals Finance Agent V2 for financial AI tasks, and Harvey's Legal Agent benchmark. It outscored Anthropic's flagship on the coding benchmark, which is timely given that, as our 1 September story on Anthropic's new Claude models reported, Anthropic has also been pushing harder performance at lower price points.
Aigora.ai CEO John Ennis told The Verge AI the result offered "Opus 5 coding quality but at a fraction of the cost and super fast." That deserves a small asterisk. CEOs announcing a cheap model matches a premium one aren't a neutral source, and real-world results vary by task.
What about the security version?
Alongside the main release, Google launched Gemini 3.8 Flash Cyber, tuned for offensive security research and not open to the public. Our 2 September story on the Fairwind Program covers the detail: access runs through a vetted group of 650 members, including cybersecurity firm CrowdStrike and the Center for Internet Security. Members can also use CodeMender, an AI agent designed to find and fix software vulnerabilities automatically. The standard 3.8 Flash ships with separate safeguards against misuse in chemical, biological and nuclear domains, plus cyber attack assistance.
What should you do?
Google AI Pro or Ultra subscribers get 3.8 Flash now, inside their existing plan, though heavier use may push toward a higher tier faster than before.
Developers paying per token should run a small test batch before switching. Check actual token counts against your 3.7 Flash baseline. A 40% cost rise on a modest workload might be worth the quality jump. On a large one, do the maths first.
