Reka
Reka develops multimodal AI models and products that work with video, images, text and audio. Its GitHub organization hosts model benchmarks, SDKs and API examples.
GitHub / reka-aiNotable people
Former people
Writing, research & talks
Reka Vision splits video understanding into three composable modules: a one-time Watch pass that stores scene, audio and text understanding as memory, natural-language Search across long activities, and Chat that resolves relative or absolute timestamps.
- Introducing Reka Flashreka.ai
Reka open-sources a research preview of Flash 3, a 21B general-purpose reasoning model with 32k context, trained from scratch and finished with RLOO. It quantizes to roughly 11GB, positions against o1-mini and QwQ-32B, and supports forced early exit from its reasoning block.
- Reka Launches Nexusreka.ai
Reka Nexus puts configurable AI workers over document search, web browsing, code execution and multimodal input, orchestrated by the 21B Reka Flash. RL with model- and rule-based rewards yields readable execution traces, and it deploys on-prem or on-device.
- Reka Flash Updatesreka.ai
Reka Flash's update adds arbitrary-resolution image handling with stronger OCR and structured output, native audio understanding inside video, 3-5 minute clips instead of one, and 128K interleaved multimodal context, though timestamp grounding is still missing.
- Multimodal With Reka Mongodbreka.ai
A walkthrough of why text-first RAG stalls on charts, tables and PDFs, and why CLIP-style embeddings do not rescue it. The pattern: use a vision model to transcribe visual structure into markdown, persist it in a vector store, then query the structured text alongside the rest of the corpus.
- Vibe Evalreka.ai
Vibe-Eval opens 269 expert-written image-text prompts with ground-truth answers, built hard enough that frontier models miss half the hard split. The report covers model-based grading versus human judgement and why difficult prompts are hard to score at all.
Reka's flagship multimodal model, trained from scratch, lands with 128K context and native image, video and audio understanding. Claims parity with GPT-4V on MMMU and wins over Claude-3 Opus on third-party blind multimodal preference tests, with on-prem and on-device deployment options.
Reka Flash is a 21B multimodal model trained from scratch, benchmarked against Gemini Pro and GPT-3.5 on MMLU, GPQA, HumanEval and vision/video suites, with a 7B Edge variant for resource-constrained deployment and RLHF'd chat versions.