LongCat-2.0
1.6T-parameter MIT-licensed agentic coding model trained on domestic chips.
About
A 1.6-trillion-parameter MoE (~48B active) agentic coding model released under the MIT license, notable as the first model of this scale trained entirely on Chinese AI accelerators — over 50,000 domestic chips, no Nvidia hardware. It carries a native 1M-token context and Meituan reports 59.5% on SWE-Bench Pro and 70.8% on Terminal-Bench 2.1 from its own testing.
The weights shipped July 5, 2026, but at trillion-parameter scale self-hosting is datacenter-class hardware territory.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- swe-bench-pro: 59.5 (Meituan-reported)
- terminal-bench-2.1: 70.8 (Meituan-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Meituan (not ByteDance) release; 1.6T MoE / ~48B active, 1M context; weights July 5, 2026; First trillion-scale model trained entirely on domestic Chinese accelerators (50k+ chips); SWE-Bench Pro 59.5% and Terminal-Bench 70.8% are Meituan-reported; independent confirmation pending. License: MIT — fully permissive, confirmed across multiple sources..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Meituan (not ByteDance) release; 1.6T MoE / ~48B active, 1M context; weights July 5, 2026
- · First trillion-scale model trained entirely on domestic Chinese accelerators (50k+ chips)
- · SWE-Bench Pro 59.5% and Terminal-Bench 70.8% are Meituan-reported; independent confirmation pending