Gemini 3.8 Flash vs 3.7 Flash: Same Price, 45% Higher Bill

Gemini 3.8 Flash costs exactly what Gemini 3.7 Flash costs. Every cell on Google’s pricing page is identical for the two models, down to the footnote that doubles both on 1 January 2027. The bill is not identical. Artificial Analysis measured 3.8 Flash spending 30% more output tokens per task, and its cost per Intelligence Index task rising from $0.40 to $0.58.

TL;DR

• Per token: same. $0.75 in, $3.75 out, $0.075 cached input, Batch and Flex at half, Priority at 1.8x, all doubling on 1 January 2027. Nothing to compare.
• Per task: 3.8 Flash costs about 45% more at high effort. $0.58 against $0.40 on Artificial Analysis, for 59 points against 56. AA’s own rounding of the same gap is “~40%”.
• The near-free upgrade is medium effort. 3.8 Flash at its default medium scores 57 for $0.41 per task, one point up on 3.7 Flash for one cent.
• Google says staying is fine. 3.7 Flash “remains fully supported for efficiency-first workloads”. That is the vendor’s sentence, not ours.

What Google Changed and What It Did Not

Gemini 3.8 Flash shipped on 2 September 2026, three weeks after 3.7 Flash on 13 August. Google’s announcement describes the difference as behaviour, not price: “3.8 Flash works harder. On complex tasks, it exhibits greater diligence — executing extra reasoning steps, and calling tools iteratively. At times, the model might use more tokens to maximize performance, especially at higher effort levels.”

Then the recommendation most write-ups skip: “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.” The developer docs repeat it: “Gemini 3.8 Flash can use more tokens on longer running and complex tasks, by design.”

So the vendor’s framing is: same price, more work per call, and an explicit off-ramp back to 3.7 Flash.

Gemini 3.8 Flash vs 3.7 Flash: Specs and Price

Prices from Google’s pricing page; index, token, cost and speed figures from Artificial Analysis, all read on 3 September 2026. Release date aside, the rows that differ are the last seven. AA attributes the cost and time rows to two things: 30% more output tokens per task, and more turns on its agentic evaluations.

How Much Better Is Gemini 3.8 Flash Than 3.7 Flash?

Gemini 3.8 Flash scores 59 on the Artificial Analysis Intelligence Index at high effort; Gemini 3.7 Flash scores 56. The three points cost about 45% more per task, $0.58 against $0.40. Artificial Analysis publishes one composite per model and effort level:

Read down the cost column and the story is plain. Gemini 3.8 Flash at high (59) costs 45% more per task than 3.7 Flash at high (56); at medium (57) it costs 2.5% more; at low (52) it costs 40% less, and 52 is a score 3.6 Flash already had.

AA’s own summary of the launch is “a 3 point Intelligence Index improvement” for “a 30% increase in average output tokens per task to 48k”. Both numbers are at high effort. At medium, the default, you get most of the improvement for almost none of the extra spend, which is the configuration this post would ship.

Gemini 3.8 Flash Token Usage: Where the Extra 30% Goes

Gemini 3.8 Flash writes about 30% more output tokens per task than 3.7 Flash at high effort, 48k on average per Artificial Analysis, and the extra tokens go into additional reasoning steps and iterative tool calls. Per-token price is the smaller half of a bill. The other half is how many tokens the model writes before it stops, and on this generation that number moved.

Over the full index, AA logged 64M output tokens for 3.7 Flash, which it describes as “fairly concise in comparison to the median of 71M”, and 120M for 3.8 Flash, “very verbose in comparison to the median”. The per-task average AA quotes is 48k for 3.8 Flash at high effort, 30% above 3.7 Flash.

The tokens go where Google said they would: extra reasoning steps and iterative tool calls. On the tasks that need them, that is the three points. On tasks that do not, it is 30% more output billed at $3.75 per million for the same answer. Which of those describes your traffic is not something a benchmark can tell you; the completion_tokens field on your own calls can.

Speed Is a Wash

3.8 Flash streams at roughly 302 output tokens per second on AA’s measurement, 3.7 Flash at 287. The 5% speed gain does not offset the 30% token increase, which is why AA’s time per task goes up, from 2.2 minutes to 2.5 at high effort, rather than down. If wall-clock latency is what you are optimising, low effort on 3.8 Flash at 0.8 minutes per task is the lever, and it costs you the score.

The Migration Is One String

Nothing in the request changes except the model name:

- model="gemini-3.7-flash"
+ model="gemini-3.8-flash"

Both models accept low, medium and high, default to medium, and reject minimal with an error; the 3.7 Flash API guide has the exact 400 response and what the cheapest legal tier costs. Context window and 64K output ceiling are unchanged. On the OpenAI-compatible surface the parameter is reasoning_effort; on Google’s native protocol it is thinking_level.

Two things to re-check after the swap:

• Output budgets and alerts. Same rate, roughly 30% more tokens per task at high effort. Anything calibrated on 3.7 Flash’s output volume will trip.
• Effort level. If you left 3.7 Flash on high because the score needed it, try 3.8 Flash on medium first. It scores one point higher than 3.7 Flash at high for a one-cent difference per AA task.

Should You Upgrade to Gemini 3.8 Flash?

• The task needs the 59, and you have checked that 3.7 Flash at 56 was actually falling short on it rather than assuming.
• You can ship at medium effort: 57 points for $0.41 per task is a strict improvement on 3.7 Flash at 56 for $0.40.
• Your traffic is agentic and tool-heavy, which is where Google says the extra steps land.

When to Stay on 3.7 Flash

• 3.7 Flash already does the job. Google’s word is “fully supported”; there is no price reason to move and a 45% per-task reason not to at high effort.
• You bill on output volume and have no headroom for a 30% increase.
• Your prompts are tuned against 3.7 Flash’s behaviour and re-tuning costs more than the point is worth.

The Cliff Both Models Share

On 1 January 2027 both models double to $1.50 input and $7.50 output. That turns 3.8 Flash’s $0.58 / $0.41 / $0.24 per-task ladder into roughly $1.16 / $0.82 / $0.48 if token counts hold, and 3.7 Flash’s $0.40 into roughly $0.80. The 45% gap between them survives the date intact; only the absolute numbers move. A budget that runs past December needs the second row.

Calling Them Through Ofox

google/gemini-3.7-flash is in the Ofox catalog at $0.75 input and $3.75 output on the Google provider route (the CloudVertex route on the same model page is $1.50 / $7.50, and that higher figure is what the pricing field in /v1/models currently reports), reachable over the OpenAI-compatible endpoint at https://api.ofox.io/v1 and over Google’s native Gemini protocol. Gemini 3.8 Flash is not in the catalog as of 3 September 2026. When it is added, the migration through Ofox is the same one-string change as above, with the google/ prefix. The live catalog at GET /v1/models is the authority on what you can call.

from openai import OpenAI

client = OpenAI(base_url="https://api.ofox.io/v1", api_key="YOUR_OFOX_API_KEY")

for model in ["google/gemini-3.7-flash"]:   # add "google/gemini-3.8-flash" once listed
    for effort in ["low", "medium", "high"]:
        r = client.chat.completions.create(
            model=model,
            reasoning_effort=effort,
            messages=[{"role": "user", "content": "Rewrite this GROUP BY as a window function."}],
        )
        print(f"{model:28} {effort:7} out={r.usage.completion_tokens:6}")

Run it on your own prompts and read the out column. The upgrade decision is the difference between the 3.7 Flash rows and the 3.8 Flash rows at the effort level you would actually ship, multiplied by $3.75 per million. Nothing in this post can give you that number; the loop can.

Sources

• https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
• https://ai.google.dev/gemini-api/docs/pricing
• https://ai.google.dev/gemini-api/docs/latest-model
• https://artificialanalysis.ai/articles/gemini-3-8-flash
• https://artificialanalysis.ai/models/gemini-3-8-flash
• https://artificialanalysis.ai/models/gemini-3-7-flash
• https://ofox.io/models/google/gemini-3.7-flash

Prices, index scores, token counts and speeds checked on 3 September 2026. Ofox catalog listing checked against the live /v1/models endpoint the same day; Ofox prices are from the model page.