Reka Flash 3

Reka Flash 3

Control the Reasoning Depth of Your Open LLM

6 followers

Reka Flash 3 is the open-source 21B LLM. You can control the reasoning depth of it! Use tags to adjust "thinking time" for complex tasks.
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Reka Flash 3 gallery image
Free
Launch Team
Anima - Vibe Coding for Product Teams
Build websites and apps with AI that understands design.
Promoted

What do you think? …

Zac Zuo

Hi everyone!

Sharing Reka Flash 3, a new open-source language model from Reka that gives you something unusual: control over how long the model "thinks" before answering!

Most LLMs have a fixed reasoning process. With Reka Flash 3, you can use <reasoning> </reasoning> tags in your prompt to tell the model to spend more time on complex problems.

It's like giving it a "thinking budget." And even with this extra thinking time, it's still fast – performance comparable to GPT-4o mini, but with a much smaller (21B parameter) and open-source (Apache 2.0) model.

Explore it yourself in Reka Space.