Financial Times · TechnologyJul 16, 2026, 02:57 UTC+8

Mira Murati's Thinking Machines debuts model, drawing from Chinese rivals

Mira Murati’s Thinking Machines draws from Chinese rivals in debut AI model

Mira Murati's startup Thinking Machines released its first model. The FT reports the model borrows training techniques from Chinese firms like DeepSeek, achieving near-frontier performance with less compute. The article does not disclose the model name, parameter count, benchmark scores, or whether it is open-weight.

Why it matters: Mira Murati's startup debut with FT exclusive on DeepSeek-inspired training — enough narrative weight for featured. But missing model name, param count, benchmarks, and open-source status keeps it at 78, not higher.

Read the original ↗Export Markdown

Take

This caught my eye because Mira Murati's Thinking Machines finally shipped its first model, and the FT says it uses training tricks similar to DeepSeek's—less compute, near-frontier results.

But the article is thin on what matters: no model name, no parameter count, no benchmark scores, no word on whether weights are open. The FT's tech reporting is generally solid, but "near-frontier performance" without MMLU, HumanEval, or SWE-bench numbers is a claim I'd discount until we see a model card.

The interesting bit is the directional signal. Murati came from OpenAI's pre-GPT-5 era, and she's publicly betting on efficiency over brute-force scaling. That's another data point in the industry's shift away from pure compute scaling. Still, right now it's a headline and a qualitative summary—worth watching, not worth citing.

Source: Financial Times · Technology · ft.com
Published: Jul 16, 2026, 02:57 UTC+8 · Jul 16, 2026