Cohere发布North Small Translate：218B
Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
垂直领域专用大模型落地典范，MoE架构兼顾性能与成本，多语言翻译能力实测领先，值得开发者关注其部署方案与性价比。
Cohere has released North Small Translate, an open-weight machine translation model from Cohere and Cohere Labs. It is a sparse Mixture-of-Experts (MoE) model with 218B total and 25B active parameters. It covers 50 languages, from Albanian to Vietnamese. On Cohere’s WMT26 evaluation, it scores 83.6 averaged across all languages. Cohere says that beats DeepL and Google Translate, plus open options like GLM 5.2 and Mistral Large 3.
Cohere 发布了 North Small Translate，这是 Cohere 与 Cohere Labs 联合推出的开源权重机器翻译模型。它是一个稀疏混合专家（MoE）模型，总参数量为 218B，激活参数量为 25B。该模型支持从阿尔巴尼亚语到越南语的 50 种语言。在 Cohere 的 WMT26 评估中，它在所有语言上的平均得分为 83.6。Cohere 表示，这一成绩超越了 DeepL 和 Google Translate，以及 GLM 5.2 和 Mistral Large 3 等开源选项。
Is it deployable? Yes. Call it free on Cohere’s API until rate limits, self-host it non-commercially, or license it commercially.
它可以部署吗？可以。你可以在 Cohere 的 API 上免费调用它（直到达到速率限制），非商业性地自行托管，或以商业许可方式使用。
Back to Where the Transformer Started
回到 Transformer 的起点
Google researchers introduced the Transformer in 2017 with Attention Is All You Need. Its main results came from WMT 2014 English-to-German and English-to-French translation. 9 years later, Cohere is returning to that original problem with a dedicated model. Cohere’s launch post on X frames translation as a sovereignty issue. Organizations that cannot communicate globally cannot stay sovereign.
Google 研究人员在 2017 年通过《Attention Is All You Need》一文引入了 Transformer。其主要成果来自 WMT 2014 英语到德语和英语到法语的翻译任务。9 年后，Cohere 带着一个专用模型回归到了这个原始问题。Cohere 在 X 平台上的发布帖将翻译视为一个主权问题。无法进行全球沟通的组织将无法保持其主权。
North Small Translate is the first translation model in Cohere’s North family. It follows Tiny Aya and Command A Translate in Cohere’s multilingual lineage. Cohere built it with RWS, whose Language Weaver scientists and language experts shaped its real-world quality.
North Small Translate 是 Cohere North 系列中的第一个翻译模型。它遵循了 Cohere 多语言谱系中的 Tiny Aya 和 Command A Translate。Cohere 与 RWS 合作构建了该模型，RWS 的语言 Weaver 科学家和语言专家塑造了其现实世界的质量。
Architecture
架构
The model structure describes a decoder-only sparse MoE Transformer. Here are the key details:
模型结构描述了一个仅解码器的稀疏 MoE Transformer。以下是关键细节：
- Experts: 128 experts, 8 activated per token, plus shared experts applied to every token.
- Router: A sigmoid over expert logits, normalized over the selected top-k.
- Attention: Sliding-window layers (window 4096, RoPE) and global layers without positional embeddings, interleaved 3:1.
- Lineage: That attention layout was first introduced in Command A.
- Context: 16K input and 16K output tokens, text only.
- Training: Post-trained specifically for translation quality.
- 专家：128 个专家，每个 token 激活 8 个，加上应用于每个 token 的共享专家。
- 路由器：对专家 logits 进行 sigmoid 处理，并在选定的 top-k 上进行归一化。
- 注意力机制：滑动窗口层（窗口大小 4096，RoPE）和不带位置嵌入的全局层，以 3:1 的比例交错排列。
- 谱系：这种注意力布局首次在 Command A 中引入。
- 上下文：16K 输入 token 和 16K 输出 token，仅限文本。
- 训练：专门针对翻译质量进行了后训练。
About 11.5% of the weights are active per token. Per-token compute tracks the 25B active parameters. Memory still has to hold all 218B.
每个 token 约有 11.5% 的权重处于激活状态。每个 token 的计算量对应于 25B 的激活参数。内存仍需容纳全部 218B 参数。
Benchmarks
基准测试
Cohere team reports these WMT26 all-languages scores in its launch blog:
Cohere 团队在其发布博客中报告了以下 WMT26 全语言得分：
| Model | WMT26 score |
|---|---|
| North Small Translate (Agentic) | 84.36 |
| North Small Translate | 83.60 |
| Qwen 3.5 397B A17B | 81.56 |
| DeepL NextGen | 81.37 |
| Gemma 4 31B (on) | 79.46 |
| GLM 5.2 FP8 | 76.50 |
| Google Translate | 68.20 |
| 模型 | WMT26 得分 |
|---|---|
| North Small Translate (Agentic) | 84.36 |
| North Small Translate | 83.60 |
| Qwen 3.5 397B A17B | 81.56 |
| DeepL NextGen | 81.37 |
| Gemma 4 31B (on) | 79.46 |
| GLM 5.2 FP8 | 76.50 |
| Google Translate | 68.20 |
The Agentic variant runs a multi-pass workflow that finds and fixes its own errors. Cohere’s scoring bands treat 80 to 100 as perfect or minor errors only. One caveat matters here. These are Cohere’s own runs, with GPT-5.6-Sol as the judge. Treat them as vendor-reported until independent WMT26 results appear.
Agentic 变体运行多轮工作流，能够发现并修复自身错误。Cohere 的评分区间将 80 至 100 分视为完美或仅存在微小错误。此处有一个需要注意的地方：这些是 Cohere 自身的测试结果，由 GPT-5.6-Sol 担任评判者。在独立的 WMT26 结果出现之前，请将其视为厂商报告的数据。
Regionally, both versions beat Gemma 4 31B (on) across Europe. On EU languages, the standard model scores 82.17 against Gemma’s 72.73. South Asia is close, at 86.16 for North against 88.04 for Gemma.
从地区来看，两个版本在欧洲均优于 Gemma 4 31B（on）。在欧盟语言上，标准模型得分为 82.17，而 Gemma 为 72.73。南亚地区差距较小，North 得分为 86.16，Gemma 为 88.04。
Speed, Long Documents and Cost
速度、长文档与成本
In Cohere’s tests, the model produced 112 output tokens per second against 81 for Gemma 4 31B. That was at low concurrency on identical hardware. At high concurrency, the figures were 39 against 30. Cohere calls this up to 1.4x higher throughput.
在 Cohere 的测试中，该模型每秒生成 112 个输出 token，而 Gemma 4 31B 为 81。这是在相同硬件上的低并发条件下测得的。在高并发下，数据分别为 39 和 30。Cohere 称其吞吐量最高提升 1.4 倍。
Long documents are a stronger point. The model scores 48.9 when translating 2 book chapters in 1 call. Google Translate scores 21.3 and Gemma 4 31B scores 19.4. Quality is measured per paragraph with xCOMET-XL.
长文档是该模型的更强项。在一次调用中翻译两章书籍时，该模型得分为 48.9。Google Translate 得分为 21.3，Gemma 4 31B 得分为 19.4。质量通过 xCOMET-XL 按段落进行测量。
In Cohere’s cost chart, the model scores 80.1 at $0.000676 per task, averaging 661 tokens. Gemini 3.1 Pro Preview (high) costs $0.038928 per task, about 58x more. Qwen 3.5 397B A17B costs $0.004525 and Command A+ costs $0.005158.
在 Cohere 的成本图表中，该模型每任务得分 80.1，费用为 $0.000676，平均处理 661 个 token。Gemini 3.1 Pro Preview (high) 每任务费用为 $0.038928，高出约 58 倍。Qwen 3.5 397B A17B 费用为 $0.004525，Command A+ 费用为 $0.005158。
How to Run It
如何运行
The fastest path is Cohere’s Chat V2 API. The model is free there until rate limits:
最快的路径是使用 Cohere 的 Chat V2 API。该模型在此处免费，直到达到速率限制：
from cohere import ClientV2
co = ClientV2(api_key="<YOUR_API_KEY>")
response = co.chat(
model="north-small-translate-1-0",
messages=[{"role": "user",
"content": "Translate everything that follows into French:\n\nEnterprises need accurate translations of business-critical documents."}],
)
print(response.message.content[0].text)from cohere import ClientV2
co = ClientV2(api_key="<YOUR_API_KEY>")
response = co.chat(
model="north-small-translate-1-0",
messages=[{"role": "user",
"content": "Translate everything that follows into French:\n\nEnterprises need accurate translations of business-critical documents."}],
)
print(response.message.content[0].text)For self-hosting, Cohere publishes 3 checkpoints, the same ones it serves in production:
对于自托管部署，Cohere 发布了 3 个检查点，与生产环境提供的一致：
| Checkpoint | Blackwell | Hopper |
|---|---|---|
| BF16 | 4x B200 | 8x H100 |
| FP8 | 2x B200 | 4x H100 |
| NVFP4 W4A16 | 1x B200 | 2x H100 |
| 检查点 | Blackwell | Hopper |
|---|---|---|
| BF16 | 4x B200 | 8x H100 |
| FP8 | 2x B200 | 4x H100 |
| NVFP4 W4A16 | 1x B200 | 2x H100 |
Key Takeaways
关键要点
- Cohere’s North Small Translate is a 218B MoE with 25B active parameters.
- It scores 83.6 on WMT26 across all languages, 84.36 in agentic mode.
- All scores are vendor-reported and judged by GPT-5.6-Sol.
- The 4-bit checkpoint runs on 1x B200 or 2x H100.
- Cohere 的 North Small Translate 是一个拥有 218B 参数的 MoE 模型，其中活跃参数为 25B。
- 它在所有语言的 WMT26 测试中得分为 83.6，在 agentic 模式下得分为 84.36。
- 所有分数均为厂商报告，并由 GPT-5.6-Sol 评判。
- 4-bit 检查点可在 1x B200 或 2x H100 上运行。
Check out the Model here. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
在此查看模型。此外，欢迎在 Twitter 上关注我们，别忘了加入我们有 15 万 + 成员的 ML SubReddit，并订阅我们的新闻通讯。等等！你在 Telegram 上吗？现在你也可以加入我们的 Telegram 群组。
Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us
需要与我们合作推广您的 GitHub 仓库、Hugging Face 页面、产品发布或网络研讨会等？请联系我们
The post Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages appeared first on MarkTechPost.
本文最初发布于 MarkTechPost。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力