Laguna S 2.1
poolside.ai
July 21, 2026
19 min read
🔥🔥🔥🔥🔥
65/100
Summary
Laguna S 2.1 has been released, enhancing model capabilities for longer horizon tasks. The update includes improvements in training methodology and post-training task distribution.
Key Takeaways
- Laguna S 2.1 is a 118 billion parameter Mixture-of-Experts model that supports a context window of up to 1 million tokens and was developed in under nine weeks.
- The model achieved a score of 70.2% on the Terminal-Bench 2.1 benchmark, making it the most capable coding model in its weight class.
- Laguna S 2.1 scored 40.4% on the DeepSWE v1.1 benchmark in thinking mode, indicating significant room for improvement in long-horizon tasks.
- Full evaluation trajectories for Laguna S 2.1 are available for public access at trajectories.poolside.ai.
Coverage & discussions
Hacker News · poolside.ai
Community discussions
Hacker News discussion ↗
Historical count: 5 · 5 saved in sample · Historical sample
Analysis ready
Community Sentiment
Positives
- Laguna S 2.1 is impressive, and its size fits achievable home hardware, making it accessible for more users.
- The smaller Laguna XS 2.1 model is a great alternative for those with limited resources, showing commitment to democratizing AI.
- The model's capabilities are competitive with top-tier models, finding insights that only previously advanced models could catch, which is a significant leap.
- 118 billion parameters and open weights are music to the ears of AI enthusiasts, hitting the sweet spot for model size and performance.
Concerns
- Many users are reporting disappointing results due to default configurations, which raises concerns about the initial testing experience.
- The need to enable 'thinking' in the model's configuration suggests that the default setup might not fully leverage its capabilities, leading to frustration.
- Skepticism exists around the accuracy of results, especially as some users have pointed out incorrect observations during testing.
