Gemini 3.8 Flash just dropped, and 305 tokens per second is hard to ignore
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
OpenAI model activity is surfacing — worth tracking for capability changes, ecosystem impact, and availability.
Gemini 3.8 Flash has been released, boasting an impressive output speed of 305 tokens per second. This significantly outperforms competitors like Muse Spark 1.2, which achieves 154 tokens per second, and GPT-5.6 Luna at 126 tokens per second. With an intelligence score of 59, Gemini 3.8 Flash is positioned closely to models scoring between 60 and 66. This speed could greatly enhance the efficiency of long coding-agent operations if it translates consistently to API usage.
The speed chart is what got me. Gemini 3.8 Flash is listed at 305 output tokens per second, almost twice the 154 shown for second-place Muse Spark 1.2 and well ahead of GPT-5.6 Luna at 126. Its intelligence score is 59, close to the group sitting between 60 and 66. If that speed holds up in normal API use, long coding-agent runs could feel a lot less painful.
Source： https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/