Gemini 3.8 Flash Ships at $0.75 per Million Input Tokens

Google DeepMind released Gemini 3.8 Flash on September 2 at $0.75 per million input tokens, plus a restricted cybersecurity variant for vetted defenders.

The release is Google's third Flash model in six weeks. In its announcement, the company said it was "building on the momentum of 3.7 Flash from three weeks ago" and called 3.8 "our best reasoning & coding model yet, at the same speed and low cost of 3.7."

Two variants, one shared core

Gemini 3.8 Flash is the general model, which Google describes as "our most intelligent workhorse model." Gemini 3.8 Flash Cyber is a security-focused build that Google calls "our most capable cybersecurity model with frontier-level performance in vulnerability detection and automated patching," available only through a new limited-access program.

Google says both builds are "powered by the same foundational intelligence" and that the coding gains came partly from "rigorous training in the highly demanding domain of cybersecurity" — the company's framing for why a security push improved general coding scores.

Pricing: flat now, double in January

3.8 Flash carries 3.7 Flash's introductory rate of $0.75 per million input tokens and $3.75 per million output tokens. A footnote sets the expiry date.

Period	Input / 1M tokens	Output / 1M tokens
Through December 31, 2026	$0.75	$3.75
From January 1, 2027	$1.50	$7.50

Google's footnote reads: "Introductory price expires on December 31, 2026. Starting January 1, 2027, $1.50/1M input tokens and $7.50/1M output tokens will apply." That is a doubling on both sides of the meter. Google did not state what 3.7 Flash will cost after the same date.

3.8 Flash vs 3.7 Flash: which should you run?

Google answered this question directly, and the answer is not "always upgrade." The company attributes 3.8's gains to a design choice: "3.8 Flash works harder. On complex tasks, it exhibits greater diligence — executing extra reasoning steps, and calling tools iteratively."

The cost consequence is stated in the same paragraph: "At times, the model might use more tokens to maximize performance, especially at higher effort levels." Per-token pricing is flat, but tokens per task can rise — the two are separate variables, and only the first is on the price sheet.

Google's own guidance for cost-sensitive deployments: "developers can utilize lower effort levels to minimize token overhead or continue to rely on Gemini 3.7 Flash, which remains fully supported for efficiency-first workloads." Teams running high-volume, single-step jobs have a documented reason to stay put.

The benchmark claims

All figures below are Google's own, published without independent verification at the time of writing:

• On DeepSWE v1.1 (Long-Horizon Software Engineering), Google says 3.8 Flash "outperforms most larger frontier models in autonomously solving complex engineering problems end to end, only at a fraction of the cost."
• 3.8 Flash scores 54.9% on HLE-Verified, which Google presents as evidence of multi-step reasoning across STEM, humanities and professional fields.
• Google says the model beats 3.7 Flash and other frontier models on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark.
• Google also claims "a significant leap in prompt injection robustness as measured by Gray Swan."

Flash Cyber's numbers are more specific

The security variant comes with tighter figures, again self-reported. On CWE-Bench, an external patching benchmark run by Collinear, Google reports a pass@1 of 47.2% against a leading frontier model's 47.8% — slightly behind on accuracy, which Google argues is offset by cost: it describes the result as sitting "on the Pareto frontier."

On an internal benchmark spanning 20 programming languages, Google says the model reaches "a success rate exceeding 70%," and that it surpasses larger frontier models on CyberGym, the standard vulnerability-discovery benchmark.

Three deployment claims are more concrete than the benchmarks:

• Google's Chrome Security team "found that 3.8 Flash Cyber produced 2.6 times more correct patches to vulnerabilities in Chrome than the best commercial models that are much larger."
• Security vendor Wiz reported "+7.5-9.7% higher recall on their internal penetration testing benchmark for a 2.3-5.2x lower cost compared to other leading frontier models."
• Google's Cloud Vulnerability Research team used the model to find "a critical foundational vulnerability in less than 2 hours, a vulnerability for which research and discovery usually takes months."

Google says it deliberately prioritised patching over offense: it "invested in vulnerability fixing from the start, and prioritized it over offensive capabilities like exploitation."

You probably cannot buy Flash Cyber

Flash Cyber ships through the Fairwind Program, announced the same day as a "limited access program for governments and trusted partners." Google says it has more than 650 participating partners globally, staged across government and national cyber authorities, critical infrastructure operators in healthcare, telecommunications, energy and financial networks, and core technology platforms.

Participation carries conditions. Google requires organisations to agree to "strict operational standards, including limiting access to employees within their internal cybersecurity, incident response, or penetration testing teams and deploying protections like multi-factor authentication."

The model is paired with Google's CodeMender harness, which the company says lets defenders "generate verified, deployment-ready patches in minutes" instead of taking weeks manually. For everyone outside the program, Google points to CodeMender with publicly available models on its Gemini Enterprise Agent Platform.

Google frames the staged rollout as a defensive head start, giving trusted defenders "a vital adaptation window to harden their systems before bad actors have a chance to exploit new capabilities."

Where 3.8 Flash is available now

Developers can reach 3.8 Flash through the Gemini API via Google AI Studio and Android Studio, and in Google Antigravity for agent-first workflows. Enterprises get it in Gemini Enterprise. On the consumer side, Google says it is live for AI Pro and Ultra subscribers across the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.

Google also says 3.8 Flash ships with safeguards covering chemical, biological, radiological and nuclear misuse and cyber offense under its Frontier Safety Framework, while Flash Cyber carries "a more permissive set of mitigations" — the stated reason it is restricted to vetted defenders.

What to watch

Two things will settle how good this release actually is. First, whether independent evaluations reproduce the DeepSWE and CWE-Bench positioning, since every number above is first-party. Second, whether the extra reasoning steps push real token bills up enough to erase the flat headline price — measurable on any production workload within a week of switching. The January 1 pricing change is the hard deadline on that math.