Macaron-V1.1
AvailableMind Lab's 752B-parameter agent model, released Sep 22 2026 with MIT-licensed open weights on Hugging Face (mindlab-research/Macaron-V1.1). Post-trained from GLM-5.3: a 744B base plus four 2B LoRA specialists for Chat, Agent, Coding and Generative UI, served with a mixture-of-LoRA harness. Follows Macaron-V1 (GLM-5.2 base); Mind Lab says the V1 to V1.1 iteration took two weeks on its Mint Recursive training platform. Reported scores: DeepSWE v1.1 71.7, SWE-Marathon 50.0, AutomationBench 53.7, ChatBench v2 66.7, UI4ABench v2 83.2 (vendor-reported). Hosted API via Mint Recursive. Active parameters and context window not stated by Mind Lab.
Specifications
- License
- Open source · MIT
- Weights
- Downloadable
- Architecture
- Mixture-of-Experts
- Parameters
- 752B
- Context window
- — tokens
- Max output
- —
- Knowledge cutoff
- —
- Price (in / out, $/M)
- —
- Modalities
- TextCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.