TLDRocket
Sign in

Mixture-of-Experts

82 summarised stories about Mixture-of-Experts, each linking back to the original source. Browse all topics →

+ Follow this topic

Wednesday, 22 July 2026

Introducing Laguna S 2.1

Poolside 2 months ago 17 ● 4 sources

Poolside AI released Laguna S 2.1, a 118-billion-parameter mixture-of-experts model designed for long-horizon coding tasks and agent work. The model achieved 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE v1.1, performing competitively against much larger models while requiring only 8 billion activated parameters per token and training in under nine weeks. The compact model enables complex agentic coding work to run on local machines, with the company publishing full evaluation trajectories to enable transparency about model behavior.

The Sequence AI of the Week #899: Inside Inkling: A Trillion-Parameter Model That Only Wakes Up 41 Billion at a Time

Substack 2 months ago 12

Inkling is a 975-billion-parameter language model that activates only 41 billion parameters per token through a routing mechanism that selects specialist modules. The model uses a sparse mixture-of-experts approach where a router selects six task-specific departments plus two general-purpose modules for each token, meaning only 4.2 percent of total capacity activates at any given time. This architecture allows the model to maintain enormous total capacity while keeping computational costs and memory requirements manageable during inference.

Poolside Releases Laguna S 2.1, an Open-Weight Agentic Coding Model Punching Above Its Weight Class on SWE-Bench Multilingual

MarkTechPost 2 months ago 10 ● 4 sources

Poolside released Laguna S 2.1, a 118B-parameter open-weight coding model that uses sparse mixture-of-experts to activate only 8B parameters per token while maintaining full model size in memory. The model scores 78.5% on SWE-Bench Multilingual, leading all published open models, and 70.2% on Terminal-Bench 2.1 with thinking enabled, outperforming much larger systems like DeepSeek-V4-Pro-Max and NVIDIA Nemotron 3 Ultra. At 4-bit quantization the model fits on a single NVIDIA DGX Spark with 128 GB memory, making it deployable on single-GPU hardware while competing with models several times its parameter size.

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.