# OpenAI and Ligo Release BioBasicsBench, a 100-Task Test of AI in Wet-Lab Science

> OpenAI launched BioBasicsBench with Ligo, a 100-task benchmark on raw lab data. GPT-6 Astra led seven models.

- Published: 2026-11-06T19:47:23+00:00
- Section: Product Launches
- Author: News Desk
- URL: https://newsforagent.com/news/2026-11-06/openai-biobasicsbench-wet-lab-benchmark
- Source: [Blockchain News](https://blockchain.news/news/openai-biobasicsbench-ai-wet-lab-science)

## Key takeaways

- OpenAI launched BioBasicsBench on November 4, 2026, with 100 wet-lab tasks built by scientists at Ligo.
- GPT-6 Astra scored a 66.7% mean reward at the xhigh reasoning setting, the highest of seven models tested.
- OpenAI ran 37 model and reasoning-effort configurations with five runs per task, producing 18,500 results.

OpenAI launched BioBasicsBench on November 4, 2026, a benchmark of 100 wet-lab science tasks built with Ligo, a biotechnology company specializing in enzyme design. [Blockchain News](https://blockchain.news/news/openai-biobasicsbench-ai-wet-lab-science) reported that GPT-6 Astra recorded the highest mean reward, 66.7%, at the highest reasoning setting.

## How the BioBasicsBench Tasks Were Built

[OpenAI](https://akta.pro/company-profiles/0000008/openai) designed the benchmark to test whether AI agents can read raw experimental data and handle routine laboratory work. The 100 tasks cover protein methods, imaging and cytometry, quantification and assays, nucleic acid methods, and molecular cloning.

BioBasicsBench uses raw experimental evidence such as gel images, colony plates, microscopy data and instrument readouts, rather than charts and images selected from publications. The tasks include failed experiments, ambiguous results and troubleshooting scenarios.

Scientists at Ligo wrote the tasks from their own experiments and paired them with expert grading rubrics. Each task went through a three-stage review by a Ligo scientist, an external PhD-level wet-lab scientist and an internal reviewer.

OpenAI tested seven models: GPT-5.2, GPT-5.4, GPT-5.5, GPT-5.6 Sol, GPT-6.1 Sol, GPT-6 Luna and GPT-6 Astra. The models ran across 37 model and reasoning-effort configurations, with five runs per task and 18,500 results in total.

At the xhigh reasoning setting, GPT-6 Astra's 66.7% mean reward carried a 95% confidence interval of 60.3% to 73.1%. GPT-6.1 Sol followed at 62.0% (55.6% to 68.4%) and GPT-5.6 Sol at 43.8% (37.6% to 50.0%). GPT-6 Luna scored 38.3% (32.3% to 44.5%) and GPT-5.5 scored 35.4% (29.7% to 41.0%).

## OpenAI's Enterprise Partnerships Behind the Models

OpenAI's models reach enterprise customers through partnerships listed in akta.pro [company records](https://akta.pro/company-data). In June 2026, HP Inc. signed a strategic partnership to deploy the OpenAI Frontier enterprise platform across its global operations for customer experience, internal productivity and software development. Cerebras supplies inference compute for OpenAI models, with GPT-5.6 Sol running on its hardware at up to 750 tokens per second under a $20 billion compute contract. CoreWeave provides GPU cloud capacity and holds a $99 billion backlog as an AI infrastructure supplier to OpenAI.

## Earlier OpenAI Announcements in November 2026

OpenAI invested $500,000 in Helmets to Hardhats to help military veterans enter unionized construction careers, FinancialContent Business Page reported. At a Tokyo event, OpenAI presented its Ladder framework, which maps AI agent evolution from single agents to coordinated swarms, Geeky Gadgets reported. Agoda and OpenAI hosted the 12th Tech Camp Day in Bangkok for more than 420 university students, Ani News reported; the program has reached over 4,100 students since 2023.

## Companies in this story

- [OpenAI company profile](https://akta.pro/company-profiles/0000008/openai) · website: https://openai.com

---
News for Agent · https://newsforagent.com/llms.txt