Model family

Reka Flash

Family overview — summarizes releases; licenses and availability belong to each release.

Maintained by Reka1 3

Reka Flash is a line of Reka language models; this record covers its open-weight releases. Reka Flash 3 (March 2025) is a 21B-parameter reasoning model that Reka says was pretrained from scratch on publicly accessible and synthetic data, then instruction-tuned and trained with reinforcement learning; it was released as open weights under the Apache License 2.0. Reka Flash 3.1 (July 2025) updated it with further reinforcement learning aimed at coding and agentic tasks, and was released with a 3.5-bit quantized version and the Reka Quant quantization library.1 2 3 4Fact reviewed Oct 2, 2026

Release notes: Reka Flash 3 announcement (external site: reka.ai)
Release notes: Reka Flash 3.1 and Reka Quant announcement (external site: reka.ai)
Model hub: Reka on Hugging Face (external site: huggingface.co)

Last reviewedOct 2, 2026Entry updated Oct 2, 2026Documented release Mar 10, 2025

Releases assessed

Each release is assessed on its own. This family has no family-wide openness label.

Releases in the Reka Flash family
Release	Weights	Tier (USASI rubric v0.2)	License	Released
Reka Flash 3.1	Weights: Public	Model-disclosure tier (USASI rubric v0.2): Open-weight	Apache-2.0	Jul 10, 2025

What it is useful for

Reka describes the open Flash models as suited to general chat, coding, instruction following, and function calling in low-latency or local deployments, and Flash 3.1 as a base for fine-tuning on agentic tasks.1 4Fact reviewed Oct 2, 2026

Organization context

Reka

Provenance and derivatives

Reka states that Reka Flash 3 was pretrained from scratch, and that Reka Flash 3.1 is an update to Reka Flash 3. Both are distributed in a Llama-compatible format so that libraries supporting Llama can run them; that is a file-format choice, not a statement that the weights derive from Llama.1 2 4

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S. headquarters

The Reka Flash models are developed and published by Reka AI, Inc., whose privacy policy states that it is headquartered in the United States and whose June 2026 announcement describes it as headquartered in San Francisco (see the Reka record). Each release is assessed on its own record.5 6 4

Assessed Oct 2, 2026

Sources

1.
Introducing Reka Flash (external site: reka.ai)
Reka · Announcement · published Mar 10, 2025 · accessed Oct 2, 2026 · evidence reviewed Oct 2, 2026
2.
RekaAI/reka-flash-3 model card (external site: huggingface.co)
Reka · Model card · accessed Oct 2, 2026 · evidence reviewed Oct 2, 2026
3.
Reka Flash 3.1 and Reka Quant (external site: reka.ai)
Reka · Announcement · published Jul 10, 2025 · accessed Oct 2, 2026 · evidence reviewed Oct 2, 2026
4.
RekaAI/reka-flash-3.1 model card (external site: huggingface.co)
Reka · Model card · accessed Oct 2, 2026 · evidence reviewed Oct 2, 2026
5.
Privacy Policy | Reka (external site: reka.ai)
Reka AI, Inc. · Official page · published Jun 2026 · accessed Oct 2, 2026 · evidence reviewed Oct 2, 2026
6.
Reka and Moonvalley Join Forces to Advance Models and Infrastructure for Physical AI (external site: reka.ai)
Reka · Announcement · published Jun 9, 2026 · accessed Oct 2, 2026 · evidence reviewed Oct 2, 2026

Email a correction Report a correction Edit this entry

This listing is not an endorsement, a safety assessment, or a federal approval.