β Catalog
Agnes 2.5 Pro Theta
AvailableAgnes AIOpen weights
Open-weights multimodal MoE from Singapore's Agnes AI, published on Hugging Face Nov 15 2026 under Apache 2.0. About 187.8B total parameters, 52 text decoder layers, 512 routed experts per MoE block with top-10 routing plus a shared expert, and a configured 262,144-token context. Takes text, images and video frames and outputs text; BF16 weights (~375.5 GB) need custom Transformers code (trust_remote_code=True), with 4x H200-class GPUs suggested as a starting point. Model card reports GPQA Diamond 92.9, LiveCodeBench v6 90.5 and AA-LCR 84.7 (vendor-reported, unverified).
Specifications
- License
- Open weights Β· Apache 2.0
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 187.8B
- Context window
- 262K tokens
- Max output
- β
- Knowledge cutoff
- β
- Price (in / out, $/M)
- β
- Modalities
- TextVisionVideoCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.