News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgwavvidReka Core

Reka Core

Multimodal language model by Reka

Reka Core is a large multimodal language model developed by the Reka Team. Released in April 2024, it processes text, images, video and audio. The model aims to match frontier systems in reasoning and human preference tests.

Share on WhatsApp

  • Processes text, images, video and audio inputs
  • Outperforms Gemini Ultra on video question answering
  • Ranked second in blind human multimodal chat evals
  • Preferred over GPT4-0613 in human text evaluations
  • Released by Reka Team in April 2024

What it is

Reka Core is the largest model in a series built by the Reka Team. It handles text alongside images, video and audio inputs. The system was trained from scratch to reason across these different media types. It targets high performance on both automatic tests and human judgments. The model represents the top tier of the Reka family, sitting above Flash and Edge variants. It is designed for complex tasks that require understanding visual and auditory context alongside written language.

How it came about

The Reka Team introduced Core in April 2024 via a technical report. This release included two smaller siblings, Reka Flash and Reka Edge. The team published details on training methods and evaluation results. The announcement positioned Core as a competitor to leading frontier models. It marked the entry of Reka into the market for powerful multimodal systems. The models were shipped for production use at the time of the report.

Inside the model

Core is a multimodal language model trained from scratch. It accepts four input types: text, images, video and audio. The architecture supports reasoning across these modalities simultaneously. The technical report notes that the model was built to deliver high value for its compute class. While specific layer counts or parameter totals are not detailed in the summary, the design focuses on integrating diverse data streams. Training involved comprehensive evaluations to ensure robustness across different task types.

What it can do

On image question answering tests like MMMU and VQAv2, Core performs similarly to GPT4-V. In blind human evaluations for multimodal chat, it ranked second, beating Claude 3 Opus. For text benchmarks such as MMLU and GSM8K, it rivals other top models. Human evaluators preferred Core over GPT4-0613 in text tasks. On video question answering using the Perception-Test benchmark, it outperformed Gemini Ultra. These results suggest strong capabilities in both visual and textual reasoning.

How to use it

The model is available through an API for production use. The Reka Team ships Core as part of its service offerings. Weights are not openly released for local deployment. Users access the model via the company’s platform. The technical report directs users to the Reka website for implementation details.

Reception and debate

The model received positive marks in blind third-party human evaluations. It was preferred over several established competitors in multimodal chat scenarios. Reviewers noted its competitive standing against GPT4-V on image tasks. The ability to beat Gemini Ultra on video benchmarks was highlighted. These results helped position Reka Core as a serious contender in the frontier model space upon its release.

Why it matters

Core demonstrated that smaller or efficiently trained models could challenge larger incumbents. Its performance on video and multimodal benchmarks set a new reference point. The release contributed to the growing field of native multimodal systems. By supporting audio and video natively, it expanded the scope of tasks language models could handle. The series influenced expectations for compute efficiency in high-performance models.

Questions people ask about Reka Core

What types of data can Reka Core process?

Reka Core is a multimodal model that accepts text, images, video and audio as inputs. It is designed to reason across these different media types simultaneously, allowing for complex interactions that go beyond simple text processing.

How does Reka Core compare to other major models?

It performs competitively with GPT4-V on image benchmarks and outperforms Gemini Ultra on video tasks. In blind human evaluations for chat, it ranked second, beating Claude 3 Opus, and was preferred over GPT4-0613 for text tasks.

Is Reka Core available for download?

No, the model is not available as open weights. It is shipped for production use via an API provided by the Reka Team. Users must access the model through the company’s platform rather than running it locally.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Reka Core yet, so the facts on this page come from the sources listed above: arxiv.org/abs/2404.12387. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.