News

Partly's new repair AI beats OpenAI, Google and Anthropic on key tasks

What's the deal? Partly, the Austin-based AI company building models for vehicle repair, has released Interpreter 3 and published benchmarks it says show the repair-trained model beating general AI systems from OpenAI, Google and Anthropic on key repair tasks.

What the tests show: Partly reports Interpreter 3 named the exact vehicle variant for 19 of every 20 cars, against 4 in 20 for the best general model, and returned the correct OEM part first for 75% of parts, ahead of Gemini 4 Argon at 63%. On repair planning, 37% of its suggestions matched work shops later added — more than twice the rate of the general models tested.

The company ran the benchmarks in November 2026, comparing Interpreter 3 with six current models including GPT-6 Astra, Gemini 4 Carbon and Claude Opus 5.5. The tests covered 250 vehicles, 257 parts and 100 repair jobs from a US shop whose data was not used to train the model.

Why it's hard: Partly says a single car model can be built in thousands of variants, and ADAS sensor calibration alone is written up in more than 25 ways. It says Interpreter has been trained over six years on millions of cars and billions of parts, referencing licensed data from more than 50 vehicle manufacturer brands.

What's the endgame? Interpreter powers PartlyRepair AI, the agentic assistant Partly launched at SEMA 2026 that automates research and admin across a repair job, from intake and damage capture to parts procurement and job close. The product is now in US pilots with some of the largest multi-site operators, while Interpreter already powers Partly's products in Europe and Australasia.

Why now? Collision operators are already applying AI to tasks such as after-hours calls, photo estimating and revenue capture. Partly's pitch is wider: its assistant checks estimates against repair procedures, compares parts orders and reconciles invoices, while leaving decisions to shop staff.

Partly raised a $50 million Series B led by DST Global in 2026 at a $500 million valuation. It has more than 220 employees and offices in Austin, San Francisco, London, Christchurch and Auckland.

The signal: "Every decision in repair depends on the exact car in front of you, which is what makes it so hard for general models," said Levi Fawcett, co-founder and chief executive officer of Partly. The US launch and pilots are a real test of whether a repair-specific model can move from benchmarks and demos to routine use at scale — the announcement alone does not establish adoption or measurable savings.

Read more: tractionnews.com

Image credit: Partly

More top stories