Google released Gemini 3.8 Flash on September 2. It is the third Flash update in six weeks. Prices remain $0.75 per million input tokens and $3.75 per million output tokens. A token is a text fragment used to measure requests and responses.
The new GPT-6 Astra has base rates about 13 times higher. Comparing them is straightforward: both companies publish a price per million tokens. Comparing quality would require separate trials, which we have not conducted.
Base rates
On the GPT-6 Astra model page, OpenAI lists $10 per million input tokens and $50 per million output tokens. Both rates have the same ratio to Flash pricing: about 13.3.
| Per million tokens | Gemini 3.8 Flash | GPT-6 Astra |
|---|---|---|
| Input | $0.75 | $10 |
| Output | $3.75 | $50 |
The table excludes caching and batch discounts, additional tool fees and Astra’s long-request surcharge. Above 272,000 input tokens, Astra doubles input and cache rates and increases the output rate by half. Google describes Flash pricing as introductory.
Why Flash 3.8 can cost more than Flash 3.7
Google’s announcement explicitly warns about consumption: on difficult tasks, Flash 3.8 takes additional reasoning steps and calls tools repeatedly. The company says this helps solve tasks but can require more tokens, particularly at higher reasoning effort.
In a coding application, a model can read files, change code, run tests and return to make corrections. The user submitted one task, but the model made several tool calls and processed their results. A short final answer does not show the volume of that work.
Keeping the price per million tokens therefore does not mean the bill stays unchanged after an upgrade. At the same rate, a larger billed volume costs more. The announcement gives no single multiplier for the increase: consumption depends on the task and settings.
Google is keeping the previous version
For applications where compute consumption is critical, Google suggests lowering reasoning effort or continuing to use Gemini 3.7 Flash, which remains supported. Developers can choose between the new model’s additional steps and the previous version’s more economical operation, without a mandatory move to 3.8.
