Laguna M.1

Poolside's open-weights coding model (April 2026, Apache 2.0): a 225B MoE (23B active) trained in-house on 30T tokens, posting strong agentic-coding results (SWE-bench Verified 65.4%, SWE-Bench Pro 46.9%) with native reasoning plus tool calling at a very low price. A promotional free endpoint exists; free-tier inputs may be used for training.

Laguna M.1 strengths

• Strong SWE-bench agentic coding
• Apache 2.0 open weights
• Native reasoning + tool calling
• Very cheap API
• Efficient 23B-active MoE

Pricing & context

Cost guide: a typical call of about 10K input + 2K output tokens costs roughly $0.003 at list prices. Worth modelling against cheaper tiers before committing high-volume traffic.

When to choose Laguna M.1

Laguna M.1 is best for agentic software engineering: codebase exploration, multi-file edits, test loops and CLI agents. If your workload is more cost-sensitive, weigh it against gpt-oss-120b (≈$0.03 input /1M) first.

Laguna M.1 FAQ

How much does Laguna M.1 cost?

Laguna M.1 is priced at $0.20 per 1M input tokens and $0.40 per 1M output tokens (public API list price), with a 262K tokens (32K max output) context window. A typical call of about 10K input and 2K output tokens costs roughly $0.003.

What is Laguna M.1 best for?

Laguna M.1 by Poolside is best for agentic software engineering: codebase exploration, multi-file edits, test loops and CLI agents.

How does Laguna M.1 pricing compare to Kimi K2.6?

Laguna M.1 input costs $0.20 per 1M tokens versus ≈$0.60 for Kimi K2.6, roughly 3.0x less expensive on input. Output is $0.40 vs ≈$2.50.

Is Laguna M.1 multimodal?

Laguna M.1 supports text.