AI Potluck
Model components / Base / pretrained models

Reka Core

Reka AI

Reka AI's multimodal foundation model that natively processes text, images, video, and audio (128K context). Founded by ex-DeepMind / Google / Meta researchers, Reka Core was originally pitched as a GPT-4V / Claude 3 Opus / Gemini Ultra rival. The company has since extended the stack with Reka Flash 3 (open-weights, 2025) and Reka Nexus (a multi-agent 'AI workforce' product). Reka Core remains API-only as the closed flagship; no verified 2026 successor to Core itself was found in this audit, though the surrounding product line has evolved.

Reka Core, frontier-class multimodal LLM announced 15 Apr 2024; 128K context. Proprietary: API/on-prem/on-device; no open weights. Instruction/multimodal-chat model. Verified live: launch page confirms API/on-prem/on-device availability and the MMMU/Claude-3-Opus claims. NOTE: '~5T tokens / 32 languages / from scratch' details are NOT on the cited launch page (they originate in Reka's separate technical report); verify against that source if retained. >12mo old; scored best-effort and flagged stale.

Openness

1 high confidence
1.0
weights
closed
data
closed(mix of public+licensed proprietary)
code
closed
license
Proprietary(API/on-prem, no weights)

Proprietary multimodal model; API/on-prem/on-device deployment but no downloadable weights. Confirmed on launch page.

Adoption

2 low confidence
2.0

Distributed via Reka API + Snowflake Cortex + Oracle Cloud Marketplace; a 2024 frontier-challenger now well behind. No disclosed usage; level 2 directional. (Snowflake/Oracle partnerships are not enumerated on the cited launch page itself.)

Capability

2 low confidence
2.0

Frontier-class for early 2024 (~GPT-4V tier) but ~2yr stale vs 2026 anchors. Headline claims CONFIRMED present on the primary launch page. Scored 2.

Unchanged since 2026-07-30 (last edited, not re-checked)