RekaReka Flash
Reka Flash 3
Flash 3 uses supervised training followed by reinforcement learning with model-based and rule-based rewards. Its public weights use Apache 2.0.
What the release provides
The model is primarily for English. Its Llama-compatible file format is a deployment choice, not a claim that the weights come from Llama. The card recommends removing earlier reasoning traces from later conversation turns.Reka Flash 3 model card
How to use it well
Practical guidance. Separate the answer from reasoning text in the application. For factual research, provide source material and inspect the supporting evidence rather than relying on the model’s stored knowledge.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | reka-flash-3 |
| Current API | Text input, structured output and reasoning; no tool calls in the reviewed API table.Reka current chat model contracts |