Macaron V1 Preview launches with 748B flagship and 50B local model variants
AdinaYakup · x · 2026-07-27
Macaron V1 Preview ships in two variants
Macaron0fficial says Macaron V1 Preview comes in two versions:
- Venti: a 748B flagship model, built from a 744B base + 4×1B LoRA specialists, post-trained on GLM-5.2.
- Tall: a 50B local-deployment model, built from a 35B base + 4×3.7B LoRA specialists, post-trained on Qwen 3.6.
The image also shows a benchmark comparison against models such as GLM 5.2, GPT 5.5, Claude Opus 4.8, Gemini 3.1 Pro, Qwen 3.7 Max, and Minimax M3 across tasks including ChatBench, LivingBench, VitaBench, PinchBench, ClawGym, SWE Verified, TerminalBench 2.1, DeepSWE, SWE Atlas QnA, and UI4ABench. The chart positions Macaron V1 as especially strong on several agentic and coding benchmarks.
Related event: Macaron V1 Preview Released in 748B and 50B Versions(2 posts)→
More from Models
- Kimi K3 goes live on Nebius with FP4 and claims 3–4x faster inference — teortaxesTex · 2026-07-27
- Opus-5 is shaping up as a daily driver, according to an early user — adonis_singh · 2026-07-27
- Macaron V1 Drops in Two Variants: 748B Flagship and 50B for Local Deployment — AdinaYakup · 2026-07-27
- User screenshot says ChatGPT and Kimi now outwrite Claude for storytelling — heyshrutimishra · 2026-07-27
- Merge Fusion says it beats the models it is built from at every price point — shensi · 2026-07-27
- A summarization pipeline improved noticeably after switching from Qwen MoE to GLM — xeophon · 2026-07-27