๐ŸŒ

Reka

Reka is the model family โ€” including Reka Flash and Reka Core โ€” built by Reka AI, a small independent lab founded by former DeepMind, Google Brain, and Meta AI researchers. It sits at #17, the least-used model in this ranking at roughly 0.4% usage share, reflecting its status as a much smaller player next to the big labs. Reka is best known for building genuinely multimodal models (text, image, video, and audio input) that hit competitive benchmark scores despite the company's small size and lower compute budget compared to Google, OpenAI, or Anthropic.

Quick facts
Made by: Reka AI
Access: Proprietary โ€” usage-based API (Reka Platform), with Reka Flash also downloadable as open weights for research/self-hosting
Context window: 128,000 tokens
Strengths: Native multimodal understanding (text, image, video, audio) from a compact model, competitive cost-to-performance ratio
Primary use case: Multimodal apps (video/image understanding) that want an alternative to the major-lab APIs at lower cost
Jump to: ExampleGetting startedBest for

Example

Reka's API is OpenAI-compatible and can be called with a plain curl request or the requests library against its chat completions endpoint.

import requests

response = requests.post(
    "https://api.reka.ai/v1/chat",
    headers={"X-Api-Key": "YOUR_API_KEY"},
    json={
        "model": "reka-flash",
        "messages": [
            {"role": "user", "content": "Explain what a REST API is in two sentences."}
        ]
    }
)

print(response.json()["responses"][0]["message"]["content"])

Getting started

Sign up on the Reka Platform for an API key, or download Reka Flash's open weights to self-host.

# 1. sign up for an API key at platform.reka.ai

# 2. call the API directly with curl
curl https://api.reka.ai/v1/chat \
  -H "X-Api-Key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"reka-flash","messages":[{"role":"user","content":"hello"}]}'

# 3. or download Reka Flash's open weights from Hugging Face to self-host
Best for: Small teams or indie builders working on multimodal features (video summarization, image Q&A) who want a cheaper alternative to Gemini or GPT-4o's multimodal pricing, without needing the absolute top-tier benchmark score.