Mira Murati's Thinking Machines Lab Releases Inkling, Its First AI Model, a 975-Billion-Parameter Open-Weight System

Overview

Thinking Machines Lab, the AI startup founded by former OpenAI chief technology officer Mira Murati, released its first in-house model on July 15, 2026, calling it Inkling in the company’s own announcement. The company describes Inkling as “the first in a family of models of different sizes,” a mixture-of-experts system built to be downloaded, modified, and fine-tuned by outside developers rather than accessed only through a metered API, according to Thinking Machines Lab.

Inkling carries 975 billion total parameters, of which roughly 41 billion are active for any given task, according to TechCrunch. Per the company’s own technical writeup, “each MoE layer contains 256 routed experts and 2 shared experts, with 6 routed experts active per token,” and the model interleaves “sliding-window and global layers at a 5:1 ratio with 8 KV heads,” according to Thinking Machines Lab. The model was pretrained on 45 trillion tokens of text, image, audio, and video, according to TechCrunch, and supports a context window of up to 1 million tokens, according to Silicon Republic. Its full weights are hosted on Hugging Face, “both as the original checkpoint and as an NVFP4 checkpoint for efficient inference on NVIDIA Blackwell systems,” according to Thinking Machines Lab.

What We Know

Thinking Machines Lab was “founded last February by Murati and a cohort of other former OpenAI employees,” according to Fortune. The company trained Inkling entirely on Nvidia’s GB300 NVL72 systems, according to TechCrunch — the same GB300 infrastructure the company previously expanded access to through a multibillion-dollar Google Cloud deal — under a partnership with Nvidia announced in March, according to SiliconANGLE. TechCrunch reported that Thinking Machines “struck a partnership with Nvidia in March to deploy a gigawatt of Vera Rubin computing capacity.” The company now employs “roughly 200 people,” TechCrunch reported, after “two co-founders who left for OpenAI.” A $50 billion fundraising round “was said to be coming together last November but had stalled by January,” according to TechCrunch, following an earlier raise of “$2 billion at a $12 billion valuation,” according to Fortune.

Thinking Machines Lab is candid about where Inkling stands competitively. “Inkling is not the strongest overall model available today, open or closed,” the company wrote in its own announcement, according to Thinking Machines Lab. Even so, on the Artificial Analysis Intelligence Index, Inkling “ranks three points above the previous leader, Nemotron 3 Ultra at 38,” making it “the leading open-weights model from a U.S. lab,” according to The Decoder. On the GDPval-AA v2 agent benchmark, Inkling reached an Elo rating of 1,238, beating Kimi K2.6 at 1,190 and DeepSeek v4 Flash max at 1,189, The Decoder reported — though the outlet’s own framing places Inkling as a model that “leads U.S. open models but trails China’s best” overall.

Thinking Machines emphasized efficiency over raw benchmark supremacy: “Inkling reaches a given score at fewer tokens — for example, it matches Nemotron 3 Ultra on Terminal Bench 2.1 at roughly a third of the tokens,” according to Thinking Machines Lab — a claim TechCrunch corroborated separately, reporting that “Inkling uses a third as many tokens as Nvidia’s Nemotron 3 Ultra … to hit the same coding performance,” and SiliconANGLE, which put it as Inkling achieving “comparable coding performance with Nvidia’s Nemotron 3 Ultra model, despite using two-thirds fewer tokens.” On the company’s own published benchmark table, Inkling scored 77.6% on SWE-bench Verified, 87.2% on GPQA Diamond, and 91.4% on VoiceBench, according to Thinking Machines Lab.

The company also built Inkling with a specific stance on contested topics. “We trained Inkling to answer directly on topics that may be subject to censorship,” Thinking Machines wrote, according to Thinking Machines Lab.

Alongside the flagship model, the company released a preview of a smaller sibling: “Alongside Inkling we are sharing a preview of Inkling-Small, a 276B-parameter Mixture-of-Experts model (12B active, vs. 41B for Inkling) with a different performance/latency trade-off,” according to Thinking Machines Lab. Both models are being made available through Tinker, the company’s fine-tuning platform: “Inkling is available on Tinker today with context length options of 64K and 256K tokens. We are offering Inkling at a 50% discount for a limited time,” according to Thinking Machines Lab. Tinker’s existing customers include the hedge fund Bridgewater Associates, according to Fortune and TechCrunch, which reported that a Bridgewater-related 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.”

What We Don’t Know

Thinking Machines has not said whether or when the stalled $50 billion valuation round might close, nor has it detailed Inkling’s training cost or disclosed a roadmap for the rest of the model family beyond Inkling and the Inkling-Small preview. It’s also unclear how enterprise customers beyond Bridgewater will use Tinker to customize Inkling, or how the model will perform outside the benchmark suites the company itself selected for its announcement.

Analysis

Inkling arrives as a bet that customizability, not leaderboard supremacy, is what enterprises actually want from an open-weight model — a contrast with Thinking Machines’ large, closed-model rivals. By training on the same Nvidia GB300 NVL72 hardware used elsewhere in the industry and then publishing token-efficiency comparisons against Nemotron 3 Ultra rather than claiming outright superiority, the company is positioning Inkling as a practical middle ground: strong enough to lead U.S. open-weight labs on the Artificial Analysis Intelligence Index, per The Decoder, but explicitly marketed around fine-tuning through Tinker rather than raw capability.