Reka AI iconReka AI

Multimodal AI lab behind Reka Flash and the Reka Edge vision model, with video-understanding products and APIs from $0.10 per 1M tokens

$0.10 Per 1M Input Tokens
$110M Raised Jul 2025
2022 Founded

Overview

Reka AI builds multimodal models that understand images, video and text, and sells them through an API and video products. Its developer catalog lists two models: Reka Flash 3 (64K context) and Reka Edge 2603, a 7B vision-language model released in March 2026 for robots, cameras and other on-device "physical AI" uses (16K context; image, video and text in, text out). Reka Flash 3.1 (21B) has open weights under Apache 2.0 on Hugging Face, while Reka Edge ships under a source-available Reka Edge BSL 1.1 license. Around the models Reka sells Reka Vision for video search and analysis (by API, MCP server or app), plus research, data (Claru) and inference services. The company was founded in 2022 by researchers including Dani Yogatama (CEO) and Yi Tay, and raised $110 million in July 2025 in a round led by Nvidia and Snowflake that valued it at about $1 billion. Reka Core and Reka Spark, named on older pages, are not in the current model catalog.

The Verdict

Who Should Use Reka AI?

Best For

  • Video search and analysis through Reka Vision
  • On-device vision for robots and cameras (Reka Edge, 7B)
  • Cheap multimodal API calls ($0.10 per 1M input)
  • Teams that want an Apache 2.0 21B model to self-host (Flash 3.1)

Not Ideal For

  • Long-context work (64K on Flash 3, 16K on Edge)
  • Buyers who want independent benchmarks and many user reviews
  • Commercial self-hosting of Edge without reading its BSL license

What's Great

  • Very low API prices: Flash 3 $0.10 in, $0.20 out; Edge $0.10 both ways
  • Strong backing: Nvidia and Snowflake led the 2025 round
  • Open Apache 2.0 weights for Reka Flash 3.1
  • Video products available by API, MCP and app

Watch Out For

  • Short context windows compared with frontier models
  • Reka Edge is source-available (BSL 1.1), not open source
  • G2 says there are not enough reviews for buying insight
  • No independent benchmarks found; Edge results are vendor-reported

Pricing

View all features & details

Models

  • Reka Flash 3 (API, 64K context)
  • Reka Flash 3.1: 21B, Apache 2.0 weights (Jul 2025)
  • Reka Edge 2603: 7B, image, video and text input (Mar 2026)
  • Reka Inverse Dynamics Model on Hugging Face (2026)

Products

  • Reka Cloud
  • Reka Vision: video infrastructure via API, MCP and app
  • Research (Labs), Claru (data) and Infer

Company

  • Founded 2022; CEO Dani Yogatama
  • $110M round led by Nvidia and Snowflake, about $1B valuation (Jul 2025)
  • Headquarters not stated on the official site

Real-World Usage

Backers

  • Nvidia and Snowflake led the July 2025 round
  • Valuation about $1 billion

Reviews

  • G2 lists Reka but has too few reviews to rate it
  • No independent benchmark of Reka Edge found as of Nov 2026

How It Compares

Feature Reka Google Gemma 4 MiniMax
Cheapest input per 1M $0.10 Free weights $0.15 (M3.1-Flash)
Max context 64K 256K 1M (M3)
Smallest text or vision model Edge, 7B E2B M3.1-Flash, 220B
Open license Apache 2.0 (Flash 3.1), BSL (Edge) Apache 2.0 Custom community license
Video product Video understanding (Reka Vision) None Video generation (H3)

Sources: Reka AI · Models and prices · Hugging Face · Reka Edge (Mar 2026) · Funding (Jul 2025)

Is this your tool? Add a badge