LongCat-Next
Meituan · Undisclosed (discrete native multimodal)
Making vision and speech mother tongues
Unified discrete tokens for understanding and generation: OmniDocBench 0.152 beats dedicated VL models, MathVista 83.1, GenEval 84.44 — with little “multimodal tax” on text.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- LongCat-Next89.2
Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.
Public benchmarks
as reported, not converted| Benchmark / metric | Value | Note |
|---|---|---|
| OmniDocBench | 0.152 | Lower is better; beats Qwen3-VL |
| MathVista | 83.1 | Visual math |
| GenEval | 84.44 | Image generation |
| C-Eval | 86.8 | No notable text regression |
Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.
Scenario mapping
five stress tests, one scale| Scenario | Score | Observation |
|---|---|---|
| Chinese receipt OCR | 93 · Outstanding on paper-grade documents | |
| Chart data extraction | 90 · Good on academic charts | |
| 30-min video QA | 85 · Still early | |
| Mobile GUI control | 86 · Surprisingly strong tool use | |
| Chinese cultural context | 88 · General training data |
Strengths and limitations
Strengths
- Unified architecture where understanding and generation reinforce each other.
- OmniDocBench beats dedicated VL models; the discrete-vision ceiling is broken.
- Open and auditable — a good domain foundation.
Limitations
- Early productization; APIs and docs immature.
- Video and agent capabilities still incomplete.
- High compute barrier for self-hosting.
Price and real cost
List price
¥2.5/M tokens
Open self-host, est. ¥2.5 / 1M output. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥1.5 (assuming input at 25% of output price).
Use your volumeAbout LongCat-Next
Is LongCat-Next open source?
Licence: Open (research license). Weights are available, but check the exact terms and filing requirements before commercial use.
How much does LongCat-Next cost?
Open self-host, est. ¥2.5 / 1M output. It ranks #9 with a composite of 89.2. Real bills depend on resolution, tiling, caching and volume discounts.
What is LongCat-Next best for?
Multimodal architecture research; Custom industry foundation models; Unified document understanding and generation。
Why does LongCat-Next score 89.2?
The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Document parsing & Chinese OCR; its weakest is Ecosystem, cost & availability.
Personal ranking · Not an official score
Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.