Agnes AI releases open-weight Agnes 2.5 Pro Theta, a 188B multimodal model

Nov 15, 2026
The Apache 2.0 mixture-of-experts model scores 92.9 on GPQA Diamond and 64.1 on SWE-bench Pro, according to the company's own model card.

Agnes AI has released Agnes 2.5 Pro Theta, a multimodal mixture-of-experts model with open weights, on Hugging Face under the Apache 2.0 licence. The checkpoint has about 187.8 billion total parameters, a configured context limit of 262,144 tokens, and comes with BF16 weights for local inference through Transformers, plus its image and video processors.

According to the model card, it scores 92.9 on GPQA Diamond, 90.5 on LiveCodeBench v6, 84.7 on AA-LCR and 64.1 on SWE-bench Pro. In the card's comparison table it has the highest score on five of the eight benchmarks shown against Nemotron 3.5 Super VL 120B, Qwen3.8-27B, GLM-5.3-Flash and DeepSeek-V4.1-Flash, while DeepSeek-V4.1-Flash scores higher on SciCode, HLE-Diamond and GDP.pdf.

The figures are the company's own, and the card notes that the context limit is a configuration setting rather than a claim that every context length has been benchmarked. Agnes AI says the model targets conversational reasoning, scientific problem solving, coding and visual understanding.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Daily AI Brief.

Also, consider following us on social media:

AI Brief AI Brief (X)
Free newsletter

Daily AI Brief

Daily report covering major AI developments and industry news, with both top stories and complete market updates

Guide
AI app builders

AI app builders

Tools that turn a description into a working web or mobile app, and the platforms companies use to build internal tools with AI

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more

You may also like

Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.