以巴黎为基地的Mistral 发布"Le Chonk"，一种仍然跟随中国整体的强大编码模式。

以巴黎为基地的Mistral 发布"Le Chonk"，一种仍然跟随中国整体的强大编码模式。

关键点

- Mistral拥有"大4"("Le Chonk")的重量。 该模型约有一万亿个参数，其中活性参数为520亿个，可以在经过修改的麻省理工学院许可证下在自己的硬件上运行。

- 在独立的基准测试中，该模型在编码方面表现良好。 然而在人工分析"智能指数"上，它得分为26分，落后于DeepSeek V4.1 Flash和GLM-5.3等开放的中国模型，远远落后于从Anthropic和Google的领先系统。

其优点是速度为每秒161个令牌，API价格较低，有害网络请求的拒绝率为96.6%.

Mistral拥有发布为Mistral大4的重量，其最大型号至今。 模特儿称公司在编码时打败了公开对手。 在第一次独立测试中，结果速度快而相当便宜，但在总排名中却跟随了中国模式。

在公开预览后三周，Mistral将"大4"重放于Hugging Face上。 该型号的外号是"Le Chonk". 它以文本和图像作为输入和输出文本和代码。 其总参数约为一万亿，每步活性为520亿。 自己运行它需要很多硬件。 根据推断框架SGLang的docs，标准版本运行于8个Nvidia B300 GPU上，更重压缩的变体运行于4个GB300 GPU上。

自预览以来，Mistral一直以强化学习来训练模型。 称公司带来了巨大的收益，特别是在编码和网络防御方面。 Mistral研究人员皮埃尔·斯托克和科伦廷·佩蒂特在一次新闻呼声中走过这一过程。 团队在内部构建的任务随着训练的进行而变得更加困难。 他们称对性能的唯一限制是团队仍然能够创建和验证一个任务有多困难。 Mistral称其为"两年赌注"，现今已偿清。

大 4 引导开口的编码模型，但以宽边径跟踪 Claude

经过三个多星期的训练，"大4"在"终端-奔驰4.0"的分数从28%上升到了41%. 也就是说它比GLM5.3领先40% Mistral 称大调4达到在人工分析"编码"Agent指数上为48%. 这实际上与GLM-5.3绑定，但大4使用47%的代币到达。 Claude Code与Claude Opus 5.5 超过的指数为62分。

在Cognition新的FrontierCode 2.0 基准测试上，"大4"以19.7%领先开放型号。 然而，Claude Opus 5.5评分为46%. 在由 Surge 运行的盲比中，测试者偏好从 GLM-5.3, Kimi K3 和 DeepSeek 更经常使用 Large 4 的代码。 与Opus 5.5相比，"大4"只赢得了22%的对决。

Mistral的比较忽略了Moonshot的"克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克克 Moonshot计划到2026年11月4日释放他们。 之后，Mistral可能没有太多的领先优势了，因为Kimi K3.1 Pro 超过 Terminal-Bench 4.0以67%的价格。

在网络防御的开放模式中，大4位列第三，而不是第一

Mistral称"大4"是网络防御的主要开放模式。 最终版公司称在"人工分析"网络指数上评分51%. 索引衡量一个模型是否能够在代码库中找到，复制并补补安全孔而不会打破已存在的特性。 然而，Mistral的自有图，显示前有两款开放型号，其中DeepSeek V4.1闪电为57%,MiMo-V2.6-Pro为56%. 因此，在开放型号中，大号为4/3，仅比GLM-5.3 Flash高50%。 Kimi K3.1 号机车普洛以64%领先整个指数。

额外训练勉强把针移到这里。 根据"人工分析"，预览片得分为50分，而最终版本只高出一分。

大4更能拒绝有害的请求。 Mistral 称它平均拒绝96.6%的恶意网络请求比任何其他开放模式都要多。 它仍然应该处理合法的防御任务。 Mistral承认该型号可以使真正的攻击变得容易，不过，因此它计划在未来几个月内收紧保障措施。 橙色网络防卫等选择合作伙伴已经在测试一个特殊版本的穿透测试。 准入需要经过审查，并有合同限制。

Mistral也看到了办公室工作的进展。 称大4公司在编辑财务电子表格和搜索网页时，只在开放模型中跟踪到Kimi K3.

Mistral 削减价格，但新增是许可证捕获

通过API,4个大项每百万个输入符号成本为70美元，每百万个产出符号成本为2.10美元。 在预览期间，列表价格为1.36美元和4.18美元。 Mistral在发布的两星期内将其削减为0.68分和2.09分，因此新出价大致符合旧的折扣。 比较而言，Kimi K3.1 投入费用为2美元，产出费用为10美元。

Mistral 发布该型号在经过修改的MIT许可证下。 使用、改装和商业部署是免费的。 然而，作为商业云服务提供大4的公司需要从Mistral获得单独的许可证。 公司主要将模型投放为"主权"AI，因为它完全在欧洲受训，可以在自己服务器上运行。 如果你不想自己主机，可以在Microsoft Foundry等平台上提供。

独立测试显示"大四"速度快又便宜，但不能抓中国最好

人工分析发布版本其"情报索引"周三。 此后它以高模式评出"大4"的最终版本，评分为26分。 指数的五分之三现在依赖于那些任务或解决方案不公开的测试。 这9个测试包括终端-Bench 4.0和FrontierCode 2.0.

预览片得分为25分，因此额外三周的训练总体上只增加了一分。 人造分析称大4击败了大小相近的开放模型的平均值，其中中位数为23分。 它仍然不及中国最强的开放型号。 DeepSeek V4.1 Flash和GLM-5.3各得32分，而MiMo-V2.6-Pro获得30分。 这种差距并不是新的。 在旧指数中，预览片得分38分，GLM-5.3得分45分。

大4的优点是速度和成本。 该模型每秒生成161个令牌，比预览时的111个更高。 在"人工分析"的"头相接"比较中，只有DeepSeek V4.1 Flash和Muse Wonder更快。 指数任务平均费用为2.06美元，而预览为5.54美元。 这使得大4比GLM-5.3(5.01)和Kimi K3.1 Pro(3.68)便宜得多，但比DeepSeek V4.1 Flash(0.60)和MiMo-V2.6-Pro(0.26)贵得多。 与其他大小相近的开放模型相比，人工分析仍然考虑其高端的每桶价格。

Mistral还承诺："大4"能用到代币，现在可以检查这一说法。 该模型在全指数上产生了1.1亿个出品令牌，而预览版则产生了1.5亿个出品令牌。 这是一个很大的改进。 但可比模型的中位数是一亿，所以"大4"仍然略高于它。

指数上方的差距很大。 兹克 1Z的Gemini 4 Carbon领先了59分，领先了Claude Opus 5.5. 但迄今为止，只有选择网络安全合作伙伴才能使用Gemini 4 Carbon. 顶部开口型号为Kimi K3.1 Pro，分数为46分，但其重量尚未为发布。

Mistral仍然认为它已经走上正轨。 称团队在媒体上称，它用美国和中国实验室拥有的一小部分资源工作，并将以十倍的资本走在前列。 大4号被指作为专用Mistral型号的基地。 公司还不会评论它基础上的更小的版本。

AI News Without the Hype – 由人类治愈

订阅THE DECODER免费阅读，每周AI通讯，我们独家的"AI雷达"前沿报告每年6次，全面存档访问，访问我们的评论部分。

现在订阅

本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。

Paris-based Mistral releases "Le Chonk," a strong coding model that still trails China overall

Key Points

- Mistral has released the weights for Large 4 ("Le Chonk"). The model has about one trillion parameters, 52 billion of them active, and can run on your own hardware under a modified MIT license.

- In independent benchmarks, the model does well on coding. On Artificial Analysis' Intelligence Index, though, it scores 26 points, behind open Chinese models like DeepSeek V4.1 Flash and GLM-5.3 and far behind the leading systems from Anthropic and Google.

- Its strengths are speed at 161 tokens per second, lower API prices, and a 96.6 percent refusal rate for harmful cyber requests.

Mistral has released the weights for Mistral Large 4, its largest model yet. The company says the model beats its open rivals at coding. In the first independent test, it turns out to be fast and fairly cheap, but it trails Chinese models in the overall rankings.

Three weeks after the public preview, Mistral has put the Large 4 weights on Hugging Face. The model's nickname is "Le Chonk." It takes text and images as input and outputs text and code. It has about one trillion parameters in total, with 52 billion active per step. Running it yourself takes a lot of hardware. According to the docs for the inference framework SGLang, the standard version runs on eight Nvidia B300 GPUs, and a more heavily compressed variant runs on four GB300 GPUs.

Since the preview, Mistral has kept training the model with reinforcement learning. The company says this brought big gains, especially in coding and cyber defense. Mistral researchers Pierre Stock and Corentin Petit walked through the process on a press call. The team built tasks in-house that got harder as training went on. They said the only limit on performance is how difficult a task the team can still create and verify. Mistral calls it a "two-year bet" that's now paying off.

Large 4 leads open models on coding but trails Claude by a wide margin

After three more weeks of training, Large 4's score on Terminal-Bench 4.0 rose from 28 to 41 percent. That puts it just ahead of GLM-5.3 at 40 percent. Mistral says Large 4 hits 48 percent on Artificial Analysis' Coding Agent Index. That's practically tied with GLM-5.3, but Large 4 uses 47 percent fewer tokens to get there. Claude Code with Claude Opus 5.5 tops the index at 62 points.

On Cognition's new FrontierCode 2.0 benchmark, Large 4 leads the open models at 19.7 percent. Claude Opus 5.5, however, scores 46 percent. In blind comparisons run by Surge, testers preferred Large 4's code over code from GLM-5.3, Kimi K3, and DeepSeek more often than not. Against Opus 5.5, Large 4 won only 22 percent of matchups.

Mistral's comparisons leave out Moonshot's Kimi K3.1 Pro. Technically that's fair, since its weights aren't public yet. Moonshot plans to release them by November 4, 2026. After that, Mistral probably won't have much of a lead left, because Kimi K3.1 Pro tops Terminal-Bench 4.0 at 67 percent.

Large 4 ranks third among open models in cyber defense, not first

Mistral calls Large 4 the leading open model for cyber defense. The company says the final version scores 51 percent on Artificial Analysis' Cyber Index. The index measures whether a model can find, reproduce, and patch security holes in a codebase without breaking existing features. Mistral's own chart, however, shows two open models ahead of it, with DeepSeek V4.1 Flash at 57 percent and MiMo-V2.6-Pro at 56 percent. That puts Large 4 third among open models, just ahead of GLM-5.3 Flash at 50 percent. Kimi K3.1 Pro leads the entire index at 64 percent.

The extra training barely moved the needle here. According to Artificial Analysis, the preview scored 50 points, and the final version is only one point higher.

Large 4 does better at refusing harmful requests. Mistral says it rejects 96.6 percent of malicious cyber requests on average, more than any other open model. It's still supposed to handle legitimate defensive tasks. Mistral admits the model could make real attacks easier, though, so it plans to tighten safeguards in the coming months. Select partners like Orange Cyberdefense are already testing a special version for penetration testing. Access requires a review and comes with contractual restrictions.

Mistral also sees progress on office work. The company says Large 4 trails only Kimi K3 among open models at editing financial spreadsheets and searching the web.

Mistral cuts the price but adds a license catch

Through the API, Large 4 costs $0.70 per million input tokens and $2.10 per million output tokens. During the preview, the list price was $1.36 and $4.18. Mistral cut it to $0.68 and $2.09 for two weeks at launch, so the new price roughly matches that old discount. For comparison, Kimi K3.1 Pro costs $2 for input and $10 for output.

Mistral released the model under a modified MIT license. Use, modification, and commercial deployment are free. Companies that offer Large 4 as a commercial cloud service, however, need a separate license from Mistral. The company mainly pitches the model as "sovereign" AI, since it was trained entirely in Europe and can run on your own servers. If you don't want to host it yourself, it's available on Microsoft Foundry, among other platforms.

Independent tests show Large 4 is fast and cheap but can't catch China's best

Artificial Analysis released version 5 of its Intelligence Index on Wednesday. It has since rated the final version of Large 4 in high mode, which scores 26 points. Three-fifths of the index now rely on tests whose tasks or solutions aren't public. The nine tests include Terminal-Bench 4.0 and FrontierCode 2.0.

The preview scored 25 points, so the three extra weeks of training added just one point overall. Artificial Analysis says Large 4 beats the average for open models of similar size, where the median is 23 points. It still falls short of the strongest open models from China. DeepSeek V4.1 Flash and GLM-5.3 each score 32 points, and MiMo-V2.6-Pro gets 30. That gap isn't new. In the old index, the preview scored 38 points and GLM-5.3 scored 45.

Large 4's strengths are speed and cost. The model generates 161 tokens per second, up from 111 for the preview. In Artificial Analysis' head-to-head comparison, only DeepSeek V4.1 Flash and Muse Wonder are faster. An index task costs $2.06 on average with Large 4, down from $5.54 for the preview. That makes Large 4 cheaper than GLM-5.3 ($5.01) and Kimi K3.1 Pro ($3.68), but much more expensive than DeepSeek V4.1 Flash ($0.60) and MiMo-V2.6-Pro ($0.26). Compared with other open models of similar size, Artificial Analysis still considers its per-token price on the high side.

Mistral also promises that Large 4 is efficient with tokens, and that claim can now be checked. The model generated 110 million output tokens across the full index, compared with 150 million for the preview. That's a big improvement. But the median for comparable models is 100 million, so Large 4 still sits slightly above it.

The gap at the top of the index is wide. Google's Gemini 4 Carbon leads with 59 points, one point ahead of Claude Opus 5.5. So far, though, only select cybersecurity partners can use Gemini 4 Carbon. The top open model is Kimi K3.1 Pro at 46 points, but its weights haven't been released yet.

Mistral still thinks it's on track. The team said on the press call that it works with a fraction of the resources US and Chinese labs have and would be at the top with ten times more capital. Large 4 is meant to serve as the base for specialized Mistral models. The company wouldn't comment yet on smaller versions built on it.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.

Subscribe now

Auto-fetched and translated by AI Gem Box. All rights belong to the original author.