On 5 July 2026, using the first-party public repository date, Meituan LongCat announced LongCat 2.0. Meituan released a frontier-scale model trained and deployed on AI ASIC superpods rather than a conventional NVIDIA-only training stack.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 5 July 2026, using the first-party public repository date |
| Availability or weight release | 5 July 2026 as MIT-licensed weights; public files completed over the following days |
| Release type | open-weight frontier-scale agent model |
| Access | MIT-licensed BF16 weights plus FP8 and INT8 releases |
| Architecture | a 1.6T-total, roughly 48B-active MoE with LongCat Sparse Attention and n-gram embeddings |
| Maximum stated context | one million tokens |
What changed
LongCat 2.0 combines an unusually large open checkpoint with a hardware-diversification claim. Meituan reports more than 35 trillion pre-training tokens, hundreds of billions of one-million-context tokens and no unrecoverable training rollback. Its sparse-attention indexing and 135B n-gram embedding block aim to increase capacity without paying dense inference cost on every token.
The practical comparison is therefore not simply whether LongCat 2.0 has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| Terminal-Bench 2.1 | 70.8 | Unified-harness coding-agent result |
| SWE-Bench Pro | 59.5 | Repository-level software engineering score |
| BrowseComp | 79.9 | Agentic search result reported by Meituan |
These are release-time results, not independently reproduced guarantees. The release blog did not expose a clear publication date in the fetched page, so this archive uses the first-party Hugging Face repository creation date and labels that evidence explicitly. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
LongCat 2.0 is described as a 1.6T-total, roughly 48B-active MoE with LongCat Sparse Attention and n-gram embeddings with one million tokens of stated context. Its access position at verification time is MIT-licensed BF16 weights plus FP8 and INT8 releases. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For LongCat 2.0, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
LongCat 2.0 is a major open release for well-resourced teams and a meaningful proof point for alternative accelerator stacks. Its multi-terabyte files and custom runtime requirements make deployment capability part of the adoption gate.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: Meituan LongCat official model card via Hugging Face. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
