No items found.

Blog

AllResearchEngineering
Cracked salt flat in Death Valley, California, stretching to low mountains under a pale sky, in black and white. Photo by Carol M. Highsmith, Library of Congress.Cracked salt flat in Death Valley, California, stretching to low mountains under a pale sky, in black and white. Photo by Carol M. Highsmith, Library of Congress.

Few Word Do Trick

few-word-do-trick
Nov 7, 2026Research
Arnhav Datar, Varun Prashant Gangal, Evan Jolley
Mount Blackburn, Wrangell Mountains, Alaska. Photo: National Park Service, Alaska Region (public domain)Mount Blackburn, Wrangell Mountains, Alaska. Photo: National Park Service, Alaska Region (public domain)

Introducing GlobeBench: A Benchmark for Text-Based World Models

globebench
Nov 5, 2026Research
Nicholas Saban, Charvi Bannur, Mariya I. Vasileva, Darshan Deshpande
Valley of Ten Thousand Smokes, Alaska. Photo: National Park Service (public domain)Valley of Ten Thousand Smokes, Alaska. Photo: National Park Service (public domain)

We Built Our Own Post-Training Stack to Measure What Our Tasks Are Worth

post-training-stack
Oct 29, 2026Engineering
Zhe Li, Evan Jolley, Bartosz Mielczarek, Yoshinari Fujinuma, Abdelrahman Madkour, Tobi Akomolede
Lower Falls of the Yellowstone, Wyoming. Photo: NPS, Yellowstone National Park (public domain)Lower Falls of the Yellowstone, Wyoming. Photo: NPS, Yellowstone National Park (public domain)

Introducing SpeedrunBench: A Challenging Benchmark for Frontier Agents

introducing-speedrunbench-a-challenging-benchmark-for-frontier-agents
Sep 3, 2026Research
Yoshinari Fujinuma, Keisuke Kamahori, Ryuto Koike, Abdelrahman Madkour, Varun Prashant Gangal, Monty Bichouna, Martyna Markiewicz, Shivani Jain, Duncan Curtis, Rebecca Qian, Anand Kannappan
Corona Arch, Moab, Utah. Photo: Vulturesong (CC0)Corona Arch, Moab, Utah. Photo: Vulturesong (CC0)

Introducing FigmaTrace: A Comprehensive Training Dataset for Figma Design Workflows

figmatrace-a-comprehensive-training-dataset-for-figma-design-workflows
Aug 20, 2026Research
Darshan Deshpande, Yoshinari Fujinuma, Martyna Markiewicz, Devanshu Bansal, Shivani Jain, Nicholas Saban, Chirag Maheshwari, Anand Kannappan
Supercell over Vrana Lake, Croatia. Photo: Šime Barešić (CC0)Supercell over Vrana Lake, Croatia. Photo: Šime Barešić (CC0)

Getting GLM-5.2 NVFP4 Post-Training Off the Ground

getting-glm-5-2-nvfp4-post-training-off-the-ground
Jul 8, 2026Engineering
Yoshinari Fujinuma, Zhe Li, Varun Prashant Gangal, Mariya I. Vasileva
Alabama Hills and the Sierra Nevada, California. Photo: Bureau of Land Management, California (public domain)Alabama Hills and the Sierra Nevada, California. Photo: Bureau of Land Management, California (public domain)

Introducing Generative Simulators: Autonomously Scaling Environments for Agents

introducing-generative-simulators
Dec 17, 2025Research
Darshan Deshpande, Varun Prashant Gangal, Li Siyan, Anand Kannappan, Rebecca Qian
Ice cave, Mendenhall Glacier, Alaska. Photo: Forest Service Alaska Region, USDA (public domain)Ice cave, Mendenhall Glacier, Alaska. Photo: Forest Service Alaska Region, USDA (public domain)

Introducing MEMTRACK: A Benchmark for Agent Memory

memtrack
Oct 14, 2025Research
Darshan Deshpande*, Varun Prashant Gangal*, Hersh Mehta, Anand Kannappan, Rebecca Qian, Peng Wang
Kelso Dunes, Mojave, California. Photo: NPS, Joshua Tree National Park (public domain)Kelso Dunes, Mojave, California. Photo: NPS, Joshua Tree National Park (public domain)

Introducing TRAIL: A Benchmark for Agentic Evaluation

introducing-trail-a-benchmark-for-agentic-evaluation
Jun 5, 2025Research
Chinmayee Kulkarni, Darshan Deshpande
Bald cypress swamp. Photo: Ryan Hagerty, USFWS (public domain)Bald cypress swamp. Photo: Ryan Hagerty, USFWS (public domain)

Introducing BLUR: A Benchmark for Tip-of-the-Tongue Search and Reasoning

the-blur-benchmark-browsing-lost-unformed-recollections
Apr 2, 2025Research
Anand Kannappan, Sky CH-Wang