Laguna 2.1 Launch Sparks Mac Acceleration Race

Episode 1 · Laguna 2.1 Debuts with Mac Inference Acceleration Contest (2026-07-28, 8 posts)

The Laguna S 2.1 model has been released, with the team claiming its parameter size is only 6% of DeepSeek V4 Pro, yet it achieves higher performance. To promote the edge deployment of the same-architecture Laguna XS 2.1, the team has partnered with Poolside and EigenLabs to launch the MLX.fast open competition, encouraging developers to optimize the model's inference speed on consumer-grade Apple Silicon Macs. The model currently runs on Mac, and internal tests have achieved a 36.8% speedup within 24 hours.

已确认

• Laguna S 2.1 is 6% the size of DeepSeek V4 Pro and delivers higher performance.
• Laguna XS 2.1 is already capable of running on consumer-grade Macs.
• The internal team boosted Laguna's operating speed by 36.8% within 24 hours.
• Poolside, EigenLabs, and MLX.fast launched an open competition aimed at further accelerating Laguna XS 2.1 inference on consumer-grade Macs.

为什么重要

• 小模型高性能: Achieving higher performance than large models at a tiny fraction (6%) of the size could drastically lower the compute threshold for AI inference, provided the data holds true.
• 端侧推理优化: The open-source optimization contest for Apple Silicon Macs will help drive the adoption and localized running of complex models on consumer-grade hardware.
• 基准防作弊机制: The team is enhancing the verifier harness to reduce benchmark hacking and cheating, and experimenting with an open, continuous improvement mechanism that allows participants to contribute agent progress, offering a new reference for fairness in AI performance evaluation.

• Laguna S 2.1 claims 6% of DeepSeek V4 Pro’s size with higher performance — gajesh · 2026-07-28
• Laguna says it sped up 36.8% in 24 hours and is opening benchmark improvement — gajesh · 2026-07-28
• Laguna gets 36.8% faster in 24 hours as its benchmark harness is hardened — gajesh · 2026-07-28
• Open competition targets 36.8% faster Laguna XS 2.1 inference on consumer Macs — gajesh · 2026-07-28
• Laguna S 2.1 gets a Mac-speed push with MLX.fast optimization contest — gajesh · 2026-07-28
• Poolside and EigenLabs launch an open contest to speed up Laguna XS 2.1 on Apple Silicon Macs — gajesh · 2026-07-28
• MLX.fast opens a public competition to make Laguna XS 2.1 run faster on consumer Macs — gajesh · 2026-07-28
• Poolside says its 100B open-weight model runs on a 128GB MacBook — gajesh · 2026-07-28

Episode 2 · Open Source Community Boosts Mac LLM Inference Speed by Over 80% (2026-07-30, 2 posts)

In just 24 hours, the open-source community achieved a breakthrough in the Mac inference optimization challenge for the Laguna XS 2.1 model. Without using speculative decoding, they successfully boosted inference speeds by 80.6%.

• Open Community Boosts Local LLM Inference on Mac by 80.6% Without Speculative Decoding — gajesh · 2026-07-30
• Mac Inference Speed Surges 80% as Open Source Community Breaks Performance Limits — gajesh · 2026-07-30