Gemini 3.8 Flash Price 2026: Same as 3.7 (US$0.75 in / US$3.75 out)

On 2026-09-02 Google shipped two models at once: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. That is the third Flash release in six weeks (3.6, then 3.7, now 3.8), and only three weeks after 3.7 Flash.

The first question everyone asked was whether the price went up. We read the official pricing table directly. The answer: it did not, and 3.8 is listed at exactly the same numbers as 3.7. So this piece is not about a price gap. It is about whether moving to a same-priced model actually costs you the same.

1. Pin the price table down first

From Google’s official API pricing page (ai.google.dev/gemini-api/docs/pricing, read on 2026-09-03), paid tier, standard:

The 3.7 Flash row is identical to the 3.8 Flash row — same numbers, same promotional end date. Two consequences:

1. Moving from 3.7 to 3.8 does not change your unit price. You do not need to redo the budget sheet.
2. That US$0.75 / US$3.75 pair is a promotional price that ends on 2026-12-31 and doubles at new year. We unpacked that trap when 3.7 launched: the “half price” on Gemini 3.7 Flash was a promotion.

On the free tier, the official pricing table lists input, output and context caching for 3.8 Flash as Free of charge. Rate limits still apply; the pricing page we checked does not put the specific requests-per-minute or per-day numbers in that same table, and we do not estimate them.

2. Real difference #1: the same unit price is not the same invoice

This is the part most coverage skips.

Google’s own announcement says 3.8 Flash works harder on complex tasks — it runs additional reasoning steps and iterative tool calls. And in Gemini’s billing rules, thinking tokens are billed at the output rate (the official output row literally says “including thinking tokens”).

Put those two facts together:

Same unit price x more output tokens per answer = a possibly higher bill per call than 3.7.

So “same price, switch everything over” is the wrong move. The right move is to run a comparison batch on your own workload and compare not the unit price but the total tokens spent finishing the same job. In particular:

• Agentic, multi-step workflows (the ones that call tools repeatedly): this is where the extra steps can inflate cost the most — and also where 3.8 most plausibly earns it.
• Single-turn short Q&A: usually a small delta, low-risk to switch.
• Fixed-schema extraction and classification: these tasks never needed “harder work”, so switching may just add reasoning cost for no gain.

3. Real difference #2: what the published scores actually say

The announcement gives few hard numbers. We list only the ones with an explicit value, and keep Google’s relative wording for the rest instead of inventing figures:

• HLE-Verified: 54.9% (the one concrete score in the announcement)
• DeepSWE v1.1: described as beating most larger frontier models (no citable number given)
• Vals Finance Agent V2 and Harvey’s legal agent benchmark: described as beating 3.7 Flash and other frontier models
• Google also claims a marked improvement in prompt-injection resilience on Gray Swan

Honest caveat: apart from the 54.9%, these are prose claims, not published figures. When a secondary article turns them into precise percentages, go back to the official page.

4. Real difference #3: Flash Cyber is not something you can buy

Plenty of reposts framed Gemini 3.8 Flash Cyber as “Google launches a security model”, which reads as if you could just pick it in the API. The official page is explicit:

• It targets autonomous vulnerability discovery and automated patching. Published results include beating 3.5 Flash Cyber and larger frontier models on CyberGym, over 70% success on real-world vulnerability discovery, and 47.2% pass@1 on CWE-Bench.
• Access is only through the Fairwind Program, aimed at trusted government bodies, critical infrastructure operators and software maintainers, and it requires an application.

In other words, ordinary developers and ordinary companies cannot get the Cyber variant today. What you can select in the model list is 3.8 Flash.

5. Where 3.8 Flash already ships

Channels listed in the announcement:

• Developers: Gemini API, Google AI Studio, Android Studio, Google Antigravity, Stitch
• Enterprise: Gemini Enterprise
• Consumers: the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets, for Google AI Pro and Ultra subscribers

To try it inside a free allowance first, start from AI Studio. We mapped where the free tier stops here: Google AI Studio free tier and overage pricing.

6. So, switch or not? Three sentences

1. You are a 3.7 Flash API user: yes, you can switch — but run a comparison batch on total tokens first, not on unit price. Measure it especially for multi-step agents.
2. You are a Gemini app, Pro or Ultra subscriber: this is a backend swap. You do nothing, and you are not charged more.
3. You are waiting for the security model: the Cyber variant is not open to the public, and there is no route around the Fairwind list.

The thing actually worth doing right now is different: unit prices double after 2026-12-31, for 3.7 and 3.8 alike. The levers that cut the bill before then are Batch and caching, and we wrote the mechanics up here: how to cut a Gemini API bill.

7. What we could not verify

House rule: if we could not verify it, we say so.

• Concrete free-tier rate limits for 3.8 Flash (requests per minute, requests per day): the pricing page we checked only says Free of charge. Unverified.
• How much more output token 3.8 spends versus 3.7 on average: Google offers prose (“works harder”) and no number. We do not estimate it.
• Exact scores for DeepSWE v1.1 and the other named benchmarks: presented as relative claims. Unverified.
• Fairwind Program eligibility bar and review time: not published. Unverified.
• Knowledge cutoff and context window ceiling for 3.8 Flash: not stated on the pages we checked. Unverified.

Official links

• Google announcement (2026-09-02): https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
• Gemini API pricing (every price above comes from here): https://ai.google.dev/gemini-api/docs/pricing
• Gemini API context caching docs: https://ai.google.dev/gemini-api/docs/caching

Related

• Gemini tool page (plans and quota verification log)
• The Gemini 3.7 Flash “half price” is a promotion that doubles in 2027
• Google AI Studio free tier and overage pricing