LongCat-2.0 (Meituan)
B Tier · 7.9/10
Meituan's open-source 1.6T-parameter MoE (~48B active) with native 1M-token context, MIT license -- trained entirely on domestic Chinese AI ASICs and revealed as the stealth 'Owl Alpha' model that had been topping OpenRouter
Score Breakdown
Benchmark Scores
Benchmarks for LongCat-2.0 (self-reported, third-party verification pending)
| Benchmark | Description | Score | |
|---|---|---|---|
| SWE-bench Pro | 59.5% | ||
| Terminal-Bench 2.1 | 70.8% | ||
| SWE-bench Multilingual | 77.3% |
Last updated: 2026-07-05
The Good and the Bad
What we like
- +Genuinely open frontier scale: 1.6T total / ~48B active per token under MIT -- the largest permissively-licensed model released to date
- +Native long context: LongCat Sparse Attention trained on hundreds of billions of tokens of 1M-context data, not a post-hoc context extension
- +Self-reported coding results in frontier territory: SWE-bench Pro 59.5 (above GPT-5.5's 58.6 and Gemini 3.1 Pro's 54.2 on the same table), Terminal-Bench 2.1 70.8
- +Proof that the domestic-Chinese-silicon stack works at frontier scale: full 35T+-token pretraining run and deployment on AI ASIC superpods with, per Meituan, no rollbacks or irrecoverable loss spikes
What could be better
- −1.6T total parameters puts self-hosting out of reach for everyone but datacenter operators -- 'open weights' here means auditable and API-competitive, not local
- −Benchmark numbers are self-published by Meituan; independent verification (Artificial Analysis etc.) still pending, and its own table shows Claude Opus 4.8 well ahead (SWE-bench Pro 69.2, Terminal-Bench 78.9)
- −Meituan is a food-delivery giant, not a dedicated AI lab -- long-term model-line commitment and Western enterprise support are unproven
- −US/EU enterprises may face procurement friction on a Chinese-trained model regardless of the MIT license
Pricing
Self-hosted (open weights)
- ✓MIT license -- commercial use permitted
- ✓Weights on Hugging Face (meituan-longcat/LongCat-2.0)
- ✓1.6T total params -- datacenter-scale deployment only
API
- ✓Pricing as reported by VentureBeat at launch -- verify on vendor platform before building
- ✓Also available via OpenRouter (previously as stealth 'Owl Alpha')
System Requirements
Hardware needed to self-host. Min = smallest viable setup (usually heavy quantization). Max = full-precision / production-grade.
| Model variant | Min | Max |
|---|---|---|
| LongCat-2.0 (1.6T MoE, ~48B active)Not practical to self-host below datacenter scale -- consume via API or OpenRouter instead. | Datacenter multi-node GPU/ASIC cluster | ASIC superpod (Meituan's own deployment) |
Known Issues
- LAUNCH (2026-06-30): Meituan open-sourced LongCat-2.0 -- 1.6T-parameter MoE (~48B activated per token), MIT license, native 1M-token context via LongCat Sparse Attention. Trained on 35T+ tokens entirely on domestic Chinese AI ASIC 'superpods' (no Nvidia), which is the geopolitical headline. Revealed as the stealth 'Owl Alpha' model that had been leading OpenRouter rankings. Self-reported benchmarks: SWE-bench Pro 59.5, Terminal-Bench 2.1 70.8, SWE-bench Multilingual 77.3. API pricing per VentureBeat: $0.75/$2.95 per 1M tokens in/out. Weights verified live on Hugging Face (safetensors present) as of 2026-07-05Source: Hugging Face model card (huggingface.co/meituan-longcat/LongCat-2.0), VentureBeat, SiliconANGLE (2026-06-30) · 2026-06-30
Best for
Teams that want frontier-class open weights for audit, fine-tuning research, or API use at aggressive pricing -- especially agentic-coding workloads where its SWE-bench Pro showing and 1M context matter.
Not for
Anyone hoping to run it locally (1.6T params is datacenter-only), or enterprises that need vendor-grade support and independently verified benchmarks before adopting.
Our Verdict
LongCat-2.0 is the most consequential open-weights release since GLM-5.2 -- not because it beats the closed frontier (Meituan's own table shows Opus 4.8 comfortably ahead) but because of what it represents: a 1.6T-parameter MIT-licensed model trained end-to-end on Chinese ASICs, from a company best known for food delivery, quietly topping OpenRouter under a pseudonym before launch. If you consume models by API, the reported $0.75/$2.95 pricing makes it worth benchmarking against your current stack. If you're tracking the open-vs-closed race, this is the data point that says frontier-scale training no longer requires Nvidia -- or a US lab.
Sources
- Hugging Face: meituan-longcat/LongCat-2.0 model card (accessed 2026-07-05)
- VentureBeat: Meituan open-sources LongCat-2.0 (accessed 2026-07-05)
- SiliconANGLE: Meituan open-sources massive LongCat-2.0 (accessed 2026-07-05)
Explore more LongCat-2.0 (Meituan) rankings
Deeper leaderboards, benchmarks, task-specific tier lists, and status/pricing pages for LongCat-2.0 (Meituan).
The Tier List Tuesday
Weekly newsletter: tier movers, new entrants, and the VS of the week. Built from our daily AI-tool sweeps. No spam, unsubscribe anytime.
Alternatives to LongCat-2.0 (Meituan)
Llama 4 (Meta)
Meta's open-weights family -- Scout (10M context), Maverick (multimodal 400B MoE). NOTE: Meta's frontier work moved to the proprietary Muse Spark line in April 2026; Llama remains downloadable and supported but is effectively in maintenance mode
Mistral AI
European AI lab with open and commercial models -- Le Chat is now **Vibe** (May 28 2026): one agent across Work Mode + Code Mode with a VS Code extension and CLI, powered by Mistral Medium 3.5 (128B dense, 256k context, 77.6% SWE-Bench Verified). Earlier 2026 line: Small 4 (119B MoE Apache 2.0), Medium 3, Voxtral TTS
DeepSeek
DeepSeek V4 shipped 2026-04-24: V4-Pro (1.6T/49B active MoE) + V4-Flash (284B/13B active), 1M native context, Hybrid Attention Architecture, open-source on HF. Trails only Gemini 3.1 Pro on world knowledge
Gemma 4 (Google)
Google DeepMind's open-weights model family -- multimodal, 256K context, runs on edge devices
DiffusionGemma (Google)
Google DeepMind's experimental open-weights TEXT-DIFFUSION model (June 10, 2026) -- 26B MoE (3.8B active), Apache 2.0, generates 256-token blocks in parallel with bidirectional attention for up to 4x faster output (1,000+ tok/s on H100). Trades some quality vs Gemma 4 for raw speed
Qwen (Alibaba)
Alibaba's open-weights + API family -- Qwen 3.7 Max flagship GA (May 20 2026: SWE-Bench Pro 60.6%, Terminal-Bench 69.7%, GPQA 92.4%, $2.50/$7.50 per 1M with 50% promo until 6/22), Qwen3.7-Plus multimodal API (Jun 2), Qwen3.6-27B dense Apache 2.0 (beats the 397B MoE on coding from one consumer GPU)
GLM / Z.ai (Zhipu AI)
Zhipu AI's open-weights flagship -- GLM-5.2 (launched 2026-06-13) is a ~753B-parameter MoE with a 1M-token context and the new IndexShare sparse-attention architecture (~2.9x lower per-token FLOPs at 1M context), MIT licensed. Vendor benchmarks put SWE-Bench Pro at 62.1 (up from GLM-5.1's 58.4) and it tops the Artificial Analysis open-weights Intelligence Index; VentureBeat reports it beats GPT-5.5 on several long-horizon coding benchmarks at roughly 1/6 the cost. Drop-in for Claude Code / Cline / OpenCode. Still trained outside the Nvidia stack on Huawei Ascend silicon
Kimi K2.6 (Moonshot)
Moonshot's 1T-parameter MoE open-weights flagship -- Kimi K2.6 (GA 2026-04-20) is #1 open-weights on Artificial Analysis Intelligence Index v4.0 (score 54, ranked #4 overall). Native video input, 256K context, Modified MIT license
Nemotron (Nvidia)
Nvidia's open-weights family -- hybrid Mamba-Transformer MoE architecture, optimized for efficient reasoning on Nvidia hardware. Nemotron 3 Ultra (550B total / 55B active) shipped 2026-06-04 as the family flagship, joining Super (120B/12B, March) and Nano
MiniMax M3
MiniMax's coding/agent flagship -- M3 (June 1 2026): 1M-token context, MSA sparse attention (>15x decoding speedup at long context), SWE-Bench Pro 59.0%, Terminal-Bench 66.0%. OPEN WEIGHTS LIVE on HuggingFace since June 12 (~428B total / ~23B active, native multimodal, minimax-community license)
Falcon (TII)
UAE's Technology Innovation Institute open-weights family -- Falcon 3 optimized for efficient sub-10B deployment on consumer hardware
gpt-oss (OpenAI)
OpenAI's FIRST open-weight models -- gpt-oss-120b (single 80GB GPU, near parity with o4-mini on reasoning) and gpt-oss-20b (runs on 16GB edge devices). Apache 2.0. Launched 2025-08-05. gpt-oss-safeguard ships in 2026 as the safety-tuned variant
IBM Granite 4.0
IBM's enterprise-focused open-weight family -- Granite 4.0 hybrid Mamba-2 + transformer architecture (70-80% memory reduction vs pure transformer), 3B to 32B sizes, Apache 2.0. First open model family to secure ISO 42001 certification. Nano 350M runs on CPU with 8-16GB RAM. 3B Vision variant landed 2026-04-01
Arcee Trinity-Large-Thinking
Arcee AI's US-made open-weight frontier reasoning model -- launched 2026-04-01. 398B total params, ~13B active. Sparse MoE (256 experts, 4 active = 1.56% routing). Apache 2.0, trained from scratch. #2 on PinchBench trailing only Claude 3.5 Opus. ~96% cheaper than Opus-4.6 on agentic tasks
Olmo 3 (AI2)
Allen Institute for AI's fully-open frontier reasoning models -- Olmo 3 family (2025-11-20) includes 7B and 32B sizes, four variants (Base, Think, Instruct, RLZero). Apache 2.0 with fully open data + checkpoints + training logs. Olmo 3-Think 32B matches Qwen3-32B-Thinking at 6x fewer training tokens
AI21 Jamba2
AI21 Labs' hybrid SSM-Transformer (Mamba-style) open-weight family -- Jamba2 launched 2026-01-08. Two sizes: 3B dense (runs on phones / laptops) and Jamba2 Mini MoE (12B active / 52B total). Apache 2.0, 256K context, mid-trained on 500B tokens
StepFun Step 3.7 Flash
StepFun's (China) agent-focused open-weight family -- Step 3.7 Flash (May 28 2026): 198B sparse MoE vision-language model, ~11B active, 256K context, Apache 2.0, ~400 tok/s, SWE-Bench Pro 56.3. Supersedes Step 3.5 Flash (Feb 2026) as the flagship
Cohere Command A
Cohere's enterprise-multilingual flagship -- 111B params, 256K context, runs on 2x H100. 23 languages. CC-BY-NC 4.0 on weights (research / non-commercial), commercial requires Cohere enterprise contract. Follow-ups: Command A Reasoning + Command A Vision