Laguna M.1: benchmark results
Poolside's open 225B MoE (23B active) agentic coding model trained with code-execution RL, scoring 72.5% on SWE-bench Verified (April 2026). Provider: Poolside. Released 2026-04-28. Access: Open.
Aggregate Rating 1579 ± 22, rank #520 of 1656 rated models, from 55 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SpeechMap Compliance | 77.5 | % Requests Completed | 73.8 |
| Tinybird AI SQL Benchmark - Success Rate | 100 | Questions answered with a valid query within 3 retries (%) | 71.7 |
| Wolfram LLM Benchmarking Project | 52.5 | Correct Functionality (%) | 68.5 |
| Tinybird AI SQL Benchmark - First-Attempt Success Rate | 96 | Questions answered with a valid query on the first attempt ( | 65.9 |
| AI Chess Leaderboard (Continuation) | 743 | Elo | 64.2 |
| AI Chess Leaderboard (Reasoning) | 769 | Elo | 59.8 |
| THOR Finding Triage - Balanced OTS | 57.1 | Class-balanced operational triage score (%) | 50 |
| THOR Finding Triage - Critical Miss Rate | 1.8 | True positives suppressed as false positives (%, lower is be | 42.9 |
| ObviousBench | 87.5 | Answer pass³ (%) | 42.5 |
| CorpFin v2 | 58.16 | Score (%) | 36.8 |
| Vals AI Terminal-Bench 2.0 | 31.46 | Accuracy (%) | 36.4 |
| Vals AI CorpFin v2 | 58.16 | Accuracy (%) | 35.3 |
Interactive version: theaggregate.ai/model?slug=laguna-m-1 · How It Works · Data refreshed daily, snapshot 2026-10-20.