Thinking Machines Lab 发布首个自研开源模型 Inkling

Thinking Machines实验室，这家由前OpenAI首席技术官Mira Murati创立的AI初创公司，于周三上午发布了其首款自主研发的AI模型，名为Inkling：https://thinkingmachines.ai/news/introducing-inkling/。与OpenAI、Anthropic或Google的旗舰模型不同，它是开放权重的，这意味着外部开发者和公司可以下载并直接修改它。

Inkling是一个专家混合系统，总共有9750亿个参数，但在执行任何特定任务时仅调用其中的一小部分——大约410亿——这是一种常见设计，可保持超大模型的运行速度更快、成本更低。据公司自身发布的资料显示，它在45万亿个文本、图像、音频和视频的标记上进行了训练，并能够原生地跨四种媒介进行推理。不过，目前它的输出仅限于文本，包括代码、格式化的产出以及结构化数据。

这个模型是Thinking Machines实验室在一年半时间里主要在幕后构建AI基础设施后的首个公开展示。一些工作已在五月的研究预览中浮出水面：https://thinkingmachines.ai/blog/interaction-models/，内容是“交互模型”——设计用于听和说（甚至打断），而不是像典型聊天机器人那样停下来等待。它也是该初创公司核心赌注的测试，即组织可以自行适应的AI将优于目前最大实验室出售的单一模板模型。

Inkling旨在提供校准后的答案，包括标记不确定性而非猜测，并允许用户在希望以速度换效率时调节“思考努力”。公司表示，在一个基准测试中，Inkling使用的标记数量是Nvidia最新一代开放权重模型Nemotron 3 Ultra的三分之一，却能达到同样的编码性能。

Thinking Machines并不声称Inkling是最先进的。其最新博客明确指出，Inkling“并非目前最强的整体模型，无论开放还是封闭。”显然，它追求的是全面均衡的性能。

这就引出了一个问题：在它所针对的企业市场中，这款产品到底是为谁设计的。目前，Thinking Machines 更像是在将 Inkling 作为一个起点而不是成品来进行市场推广，让组织通过公司的模型定制平台 Tinker 自行微调。这也意味着确保定制安全的是客户而非 Thinking Machines（微调需要严肃的机器学习人才）。

OpenAI、Anthropic 和 Google 对 ChatGPT、Claude 和 Gemini 的做法完全不同，这些模型都是首先作为通用聊天机器人构建的，然后在其基础上增加了智能代理和自主功能。

Thinking Machines 上周发布的一篇文章：https://thinkingmachines.ai/blog/the-future-worth-building-is-human/ 显然是此次发布的背景。公司在文章中认为，由一家企业集中训练后固定下来的 AI，不如由组织自行塑造的 AI 表现得好，因为太多专业知识是特定于持有这些知识的人。

反对封闭模型的其他论点也在不断增多。微软 CEO Satya Nadella 在周日发布的一篇博客中——他的公司已在 OpenAI 和 Anthropic 投资数十亿美元——警告称，企业使用专有 AI 模型实际上是在付两次钱：https://techcrunch.com/2026/07/13/satya-nadella-has-issued-a-shocking-warning-to-companies-using-ai/：一次是订阅费用，另一次是交出嵌入在提示和修正中的业务知识，这些知识可能被吸收到未来的模型版本中。

Hugging Face CEO Clem Delangue 上周在接受 TechCrunch 采访时做出了类似预测：https://techcrunch.com/2026/07/10/hugging-faces-ceo-on-why-companies-are-done-renting-their-ai/。他表示，前沿模型将越来越多地用于实验和高价值任务，而大部分生产 AI 工作将转向私有或开源替代方案——这正是 Thinking Machines 正在构建的分工模式。

Thinking Machines 方法的最清晰论据来自最近的一项项目：https://thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks/ 与世界上最大的对冲基金桥水基金合作（顺便提一下，它并不是 Thinking Machines 的投资者）。两家公司的研究人员在现有的开源模型基础上，对其进行了进一步训练，结合桥水基金自身的金融专业知识。据称，结果在金融推理测试中得分为 84.7%，超过了顶级专有 AI 模型，同时运行成本约为其十四分之一——尽管这些结果来自两家公司自己的评估，而非独立评估。

无论如何，Thinking Machines 强调的是它达到这一点的速度。OpenAI 大约花了五年时间将其技术推向市场并显示收入，Anthropic 大约花了三年。Thinking Machines 表示，它在大约九个月内就达到了同样的水平。

有人可能会想，Inkling 是否通过竞争对手模型的输出进行训练，这种做法被称为“蒸馏：https://techcrunch.com/2026/04/30/elon-musk-testifies-that-xai-trained-grok-on-openai-models/”，在行业中引起了关注：https://techcrunch.com/2026/02/23/anthropic-accuses-chinese-ai-labs-of-mining-claude-as-us-debates-ai-chip-exports/。根据公司自己的资料，简短的答案是部分如此。Thinking Machines 从零开始对 Inkling 进行了预训练，但它表示在大型强化学习接管之前，确实使用了其他开源模型——包括 Moonshot AI 的 Kimi K2.5——来生成一些早期后训练数据。公司坚称，下一款模型将完全使用自包含的后训练。

在成本方面，Thinking Machines 更加谨慎。今年三月，它与 Nvidia 建立了合作关系，部署了 1 吉瓦的 Vera Rubin 计算能力，并完全在 Nvidia 的 GB300 NVL72 系统上训练 Inkling——但尚未说明计划如何覆盖这些成本，而且据大多数说法，收入并不是优先考虑。（据报道，一轮 500 亿美元的融资去年十一月正在进行，但到一月时已陷入停滞；公司此后拒绝谈论其融资情况。）

一个相关的问题是，Thinking Machines 的支出是否会达到 OpenAI 或 Anthropic 的规模，或者其以效率为驱动的方法意味着其经济模式看起来有所不同。换句话说，该公司的赌注可能不在于它最终会像更大的竞争对手那样花钱，而在于它根本不需要——因为一旦权重公开，任何下载它们的人都不必支付 Thinking Machines 来运行它们，而这与 OpenAI 和 Anthropic 出售的计量访问不同。公司的收入必须来自 Tinker，而不是模型本身，通过训练、微调以及现在围绕其构建的托管生态系统中的分成。

至少员工人数看起来更稳定。Thinking Machines 目前大约有 200 名员工，高于今年早些时候一波离职后的水平，包括两位在今年一月离开去 OpenAI 的联合创始人：https://techcrunch.com/2026/01/14/mira-muratis-startup-thinking-machines-lab-is-losing-two-of-its-co-founders-to-openai/

就 Thinking Machines 而言，它似乎并不热衷于像行业中很多公司那样夸大个别举动。据公司内部人士称，其文化按设计偏向连续性，而不是依赖任何一个个性。这是合理的：如果团队成员更换，对公司影响不大，因为他们从未被神化。考虑到公司自身故事中仍然与其现已成名的联合创始人的名字紧密相关，这种坚持也很了不起，无论她是否有意如此。

当您通过我们文章中的链接购买时，我们可能会赚取少量佣金：https://techcrunch.com/techcrunch-affiliate-monetization-standards/。这不会影响我们的编辑独立性。

最后机会，在 TechCrunch 创始人峰会节省高达 190 美元。加入 1,000 名各阶段的创始人和风险投资人，获取可实际应用的扩展经验和有影响力的联系。优惠截止至 6 月 26 日，太平洋时间晚上 11:59。

Anthropic 最新广告让人感到毛骨悚然：https://techcrunch.com/2026/07/14/anthropics-newest-ad-is-creeping-people-out/ Lucas Ropek：https://techcrunch.com/author/lucas-ropek/

萨蒂亚·纳德拉向使用人工智能的公司发出令人震惊的警告：https://techcrunch.com/2026/07/13/satya-nadella-has-issued-a-shocking-warning-to-companies-using-ai/ 朱莉·博尔特：https://techcrunch.com/author/julie-bort/

苹果对OpenAI商业秘密诉讼中最离奇的指控：https://techcrunch.com/2026/07/13/the-wildest-allegations-in-apples-trade-secrets-lawsuit-against-openai/ 萨拉·佩雷斯：https://techcrunch.com/author/sarah-perez/

Anthropic开始为印度本地化Claude定价，这是其继美国之后的最大市场：https://techcrunch.com/2026/07/13/anthropic-starts-localizing-claude-pricing-for-india-its-biggest-market-after-the-us/ 贾格米特·辛格：https://techcrunch.com/author/jagmeet-singh/

Meta在Instagram上移除引发争议的AI功能以应对反弹：https://techcrunch.com/2026/07/10/meta-removes-controversial-ai-feature-on-instagram-after-backlash/ 卢卡斯·罗佩克：https://techcrunch.com/author/lucas-ropek/

苹果因涉嫌商业秘密盗窃起诉OpenAI：https://techcrunch.com/2026/07/10/apple-sues-openai-over-alleged-trade-secret-theft/ 萨拉·佩雷斯：https://techcrunch.com/author/sarah-perez/

Instagram用户：这就是如何阻止Meta的AI使用你的照片：https://techcrunch.com/2026/07/09/how-to-stop-metas-ai-image-generator-from-using-your-instagram-photos/ 劳伦·福里斯塔尔：https://techcrunch.com/author/lauren-forristal/

Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, released its first in-house AI model Wednesday morning, called Inkling：https://thinkingmachines.ai/news/introducing-inkling/. And unlike the flagship models from OpenAI, Anthropic, or Google, it’s open-weight, meaning outside developers and companies can download it and modify it directly.

Inkling is a mixture-of-experts system with 975 billion total parameters, though it only draws on a fraction of that — about 41 billion — for any given task, a common design that keeps very large models faster and cheaper to run. It was trained on 45 trillion tokens of text, image, audio, and video, and reasons natively across all four, according to the company’s own release materials. For now, though, its outputs are limited to text, including code, styled artifacts, and structured data.

The model is Thinking Machines Labs’ first public proof point after a year and a half spent building AI infrastructure largely out of public view. Some of that work had already surfaced in a May research preview：https://thinkingmachines.ai/blog/interaction-models/ of “interaction models” — AI designed to listen and speak (and even interrupt) instead of stop and wait as with typical chatbots. It’s also a test of the central bet behind the startup, which is that AI that organizations can adapt for themselves will outperform the one-size-fits-all models the biggest labs currently sell.

Inkling is designed to give calibrated answers, including flagging uncertainty rather than guessing, and lets users dial “thinking effort” up or down when they want to trade for speed. On one benchmark, the company says, Inkling uses a third as many tokens as Nvidia’s Nemotron 3 Ultra — its latest generation open-weight model — to hit the same coding performance.

Thinking Machines doesn’t claim Inkling is best-in-class. Its newest blog post states explicitly that Inkling is “not the strongest overall model available today, open or closed.” What it’s evidently going for instead is well-rounded performance.

That raises the question of who, within the enterprise market it’s targeting, this product is really for. Thinking Machines is, for now, marketing Inkling less as a finished product than as a starting point, something for organizations to fine-tune themselves through Tinker, the company’s model-customization platform. This also means customers, not Thinking Machines, are responsible for making sure their customizations are safe, for example. (Fine-tuning requires serious machine-earning talent.)

OpenAI, Anthropic, and Google have all taken a very different approach with ChatGPT, Claude, and Gemini, respectively, which were all built to compete as general-purpose chatbots first, with agentic, autonomous features layered on top.

A post published by Thinking Machines last week：https://thinkingmachines.ai/blog/the-future-worth-building-is-human/ was clearly meant as the backdrop for this release. AI that’s trained centrally by one company and then set in stone, the company argued in that post, underperforms AI that organizations shape themselves because so much expertise is specific to the people who hold it.

Other arguments against closed models are gaining steam. In a blog post published Sunday, Microsoft CEO Satya Nadella — whose company has invested billions in both OpenAI and Anthropic — warned that enterprises using proprietary AI models effectively pay twice：https://techcrunch.com/2026/07/13/satya-nadella-has-issued-a-shocking-warning-to-companies-using-ai/: once in subscription costs, and again by handing over business knowledge embedded in their prompts and corrections, which can be absorbed into future model versions.

Hugging Face CEO Clem Delangue made a similar prediction：https://techcrunch.com/2026/07/10/hugging-faces-ceo-on-why-companies-are-done-renting-their-ai/ in conversation with TechCrunch last week. Frontier models, he said, will increasingly be reserved for experimentation and high-value tasks, while most production AI work shifts to private or open-source alternatives — the exact split Thinking Machines is building around.

The clearest argument for Thinking Machines’ approach came from a recent project：https://thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks/ with Bridgewater Associates, the world’s largest hedge fund (which is not, for what it’s worth, a Thinking Machines investor). Researchers from both companies took an existing open-source model and trained it further on Bridgewater’s own financial expertise. The result was said to score 84.7% on financial reasoning tests, beating top proprietary AI models, while costing roughly a fourteenth as much to run — though those results come from the two companies’ own evaluation, not an independent one.

Either way, Thinking Machines is emphasizing how quickly it got here. OpenAI took roughly five years to bring its tech to market and show revenue, and Anthropic roughly three. Thinking Machines says it did the same in about nine months.

Some will wonder whether Inkling was trained on outputs from competitors’ models, a practice known as “distillation：https://techcrunch.com/2026/04/30/elon-musk-testifies-that-xai-trained-grok-on-openai-models/” that has drawn scrutiny：https://techcrunch.com/2026/02/23/anthropic-accuses-chinese-ai-labs-of-mining-claude-as-us-debates-ai-chip-exports/ across the industry. The short answer, per the company’s own materials, is partly. Thinking Machines pre-trained Inkling from scratch, but it says it used other open-weight models — including Moonshot AI’s Kimi K2.5 — to help generate some of its early post-training data before large-scale reinforcement learning took over. The next model, the company insists, will use fully self-contained post-training instead.

On the cost side, Thinking Machines has been more guarded. It struck a partnership with Nvidia in March to deploy a gigawatt of Vera Rubin computing capacity and trained Inkling entirely on Nvidia’s GB300 NVL72 systems — but hasn’t said how it plans to cover those costs, and revenue, by most accounts, hasn’t been a priority. (A reported $50 billion fundraising round was said to be coming together last November but had stalled by January; the company has declined to talk about its funding picture since.)

A related question is whether Thinking Machines’ spending will ever reach the scale of OpenAI’s or Anthropic’s, or whether its efficiency-driven approach means the economics look different. Put another way, the company’s bet may be less that it will eventually spend like its larger rivals than that it won’t need to at all — because once weights are public, nothing obligates anyone who downloads them to pay Thinking Machines to run them, unlike the metered access OpenAI and Anthropic sell. It’s Tinker, not the model itself, where the company’s revenue has to come from, via training, fine-tuning, and, now, a cut of the hosting ecosystem built around it.

Headcount, at least, looks more settled. Thinking Machines now employs roughly 200 people, up from levels reported after a wave of departures earlier this year, including two co-founders who left for OpenAI：https://techcrunch.com/2026/01/14/mira-muratis-startup-thinking-machines-lab-is-losing-two-of-its-co-founders-to-openai/ in January.

Thinking Machines, for its part, doesn’t seem interested in playing up individual moves the way much of the industry does. According to a source inside the company, its culture, by design, favors continuity over reliance on any one personality. It makes sense: it’s less of a setback when people change teams if they were never put on a pedestal to begin with. It’s also a remarkable thing for a company to insist on, given how much of its own story is still associated with the name of its now-famous co-founder, whether she planned it or not.

When you purchase through links in our articles, we may earn a small commission：https://techcrunch.com/techcrunch-affiliate-monetization-standards/. This doesn’t affect our editorial independence.

Last chance to save up to $190 on TechCrunch Founder Summit. Join 1,000+ founders and VCs at all stages for real-world scaling insights and connections that move the needle. Savings end June 26, 11:59 p.m. PT .

Anthropic’s newest ad is creeping people out：https://techcrunch.com/2026/07/14/anthropics-newest-ad-is-creeping-people-out/ Lucas Ropek：https://techcrunch.com/author/lucas-ropek/

Satya Nadella has issued a shocking warning to companies using AI：https://techcrunch.com/2026/07/13/satya-nadella-has-issued-a-shocking-warning-to-companies-using-ai/ Julie Bort：https://techcrunch.com/author/julie-bort/

The wildest allegations in Apple’s trade secrets lawsuit against OpenAI：https://techcrunch.com/2026/07/13/the-wildest-allegations-in-apples-trade-secrets-lawsuit-against-openai/ Sarah Perez：https://techcrunch.com/author/sarah-perez/

Anthropic starts localizing Claude pricing for India, its biggest market after the US：https://techcrunch.com/2026/07/13/anthropic-starts-localizing-claude-pricing-for-india-its-biggest-market-after-the-us/ Jagmeet Singh：https://techcrunch.com/author/jagmeet-singh/

Meta removes controversial AI feature on Instagram after backlash：https://techcrunch.com/2026/07/10/meta-removes-controversial-ai-feature-on-instagram-after-backlash/ Lucas Ropek：https://techcrunch.com/author/lucas-ropek/

Apple sues OpenAI over alleged trade secret theft：https://techcrunch.com/2026/07/10/apple-sues-openai-over-alleged-trade-secret-theft/ Sarah Perez：https://techcrunch.com/author/sarah-perez/

Instagram users: Here’s how to stop Meta’s AI from using your photos：https://techcrunch.com/2026/07/09/how-to-stop-metas-ai-image-generator-from-using-your-instagram-photos/ Lauren Forristal：https://techcrunch.com/author/lauren-forristal/