Gemini 3.8 Flash arrives at $0.75 per million input tokens, and that price has an end date

Google put Gemini 3.8 Flash into the Gemini API on September 2, 2026, alongside a second model built for security work that most developers will not be able to call at all. The general model is listed at $0.75 per million input tokens and $3.75 per million output tokens. Google calls that an introductory rate and gives it a deadline: on January 1, 2027 the two figures become $1.50 and $7.50, double what the model costs before that date.

Which price to budget against

The figure to plan around is the second one. A model that costs $0.75 until December 31, 2026 and $1.50 from the next day is not a $0.75 model. It is a model with a discount attached to a date. Google notes that 3.8 Flash launches at the same introductory rate Gemini 3.7 Flash launched with, so the pattern is established enough to budget for.

The same price list gives Gemini 3.5 Flash at $1.50 per million input tokens and $9.00 per million output tokens. The January rate for 3.8 Flash matches that on input and stays below it on output. Batch and Flex requests run at half the standard rate, and Priority runs at 1.8 times it, on both sides of the introductory line.

Flash Cyber is a model you probably cannot get

The second release, Gemini 3.8 Flash Cyber, has no entry on the price list. Google distributes it through the Fairwind Program, which it describes as a limited access program for governments and trusted partners. Eligibility covers government agencies and national cyber authorities, operators of critical infrastructure in healthcare, telecommunications, energy and financial networks, and core technology platform providers. Participants must accept operational standards that include restricting access to staff on their own security, incident response or penetration testing teams.

Google says the program has more than 650 participating partners globally, and that members also receive the CodeMender harness. Alongside it the company announced more than $100 million in cybersecurity funding through Google.org, including $36 million for 35 cyber clinics that serve over 1,250 hospitals, public school districts and municipal utilities in the United States.

The numbers, and who measured them

Every figure below comes from Google's own announcement. None of it has been reproduced independently, and the announcement names no outside evaluator except Gray Swan.

• On HLE-Verified, Google reports 54.9% for 3.8 Flash.
• On DeepSWE v1.1, a long-horizon software engineering benchmark, Google says the model outperforms most larger frontier models.
• On real-world vulnerability discovery across 20 programming languages, Google reports a success rate above 70% for Flash Cyber.
• On automated patching, Google reports 47.2% for Flash Cyber on a first attempt, against 47.8% for a leading frontier model it does not name.
• On prompt injection robustness measured by Gray Swan, Google reports a significant improvement and publishes no figure.

Where the model can be called, and what is missing

Google lists three routes. Developers reach 3.8 Flash through the Gemini API in Google AI Studio and Android Studio, through Google Antigravity, and through Stitch for interface generation. Enterprises get it in Gemini Enterprise. Subscribers to Google AI Pro and Ultra get it in the Gemini app, in AI Mode in Google Search and in Gemini in Google Sheets.

The announcement states neither the context window nor the knowledge cutoff, for either model, and the price list carries neither figure either. The patching result is the one place where Google prints a competitor's number beside its own, and on that metric the smaller model trails: the argument being made there is about cost, not accuracy.