Trending Now - Innovative AI Solutions Emerge to Tackle Inference Challenges
WHO:
Gimlet Labs, a startup founded by Zain Asgar, who is a Stanford adjunct professor.
WHAT:
Nvidia's Nemotron-Cascade 2 model has 30 billion parameters and activates only 3 billion parameters at inference time.
Nemotron-Cascade 2 achieved gold medal-level performance in the 2025 International Mathematical Olympiad, the International Olympiad in Informatics, and the ICPC World Finals.
The model outperformed Nvidia's existing Nemotron-3-Nano and Nemotron-3-Super models on nearly every benchmark.
On the AIME 2025 math benchmark, the Multi-Domain On-Policy Distillation (MOPD) technique recovered teacher-level performance within 30 optimization steps.
On the ArenaHard benchmark for human preference alignment, MOPD achieved a score of 85.5 on hard prompts in 52 steps.
Gimlet Labs raised an $80 million Series A funding round.
The company claims to be the first "multi-silicon inference cloud" for running AI workloads across diverse types of hardware.
The technology developed by Gimlet Labs allows AI to run simultaneously across various chip types, including NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix.
WHERE:
The specific location of Gimlet Labs is not mentioned in the articles.
WHEN:
The funding round was reported on March 23rd, 2026.
WHY:
The report emphasizes that post-training techniques like Cascade RL and MOPD can close the performance gap on targeted domains, allowing organizations to deploy strong reasoning capabilities without high infrastructure costs.
Gimlet Labs aims to solve the AI inference bottleneck.
HOW:
The company utilizes a technology that enables AI workloads to run across multiple types of hardware simultaneously.
(UPDATED March 24, 2026 01:43 UTC)
Nvidia has unveiled its latest model, the Nemotron-Cascade 2, which boasts an impressive 30 billion parameters, activating only 3 billion during inference. This model has not only achieved gold medal-level performance in prestigious competitions like the International Mathematical Olympiad and the ICPC World Finals, but it also surpassed Nvidia's previous models on nearly every benchmark. The report highlights that innovative post-training techniques, such as Cascade RL and Multi-Domain On-Policy Distillation, can significantly enhance performance in targeted areas, making advanced reasoning capabilities more accessible and cost-effective for organizations.
--
(UPDATED March 23, 2026 19:49 UTC)
Gimlet Labs, a startup founded by Stanford adjunct professor Zain Asgar, has successfully raised an impressive $80 million in a Series A funding round, as reported today. The company claims to be the first "multi-silicon inference cloud," designed to tackle the AI inference bottleneck by allowing AI to run simultaneously across various chip types, including NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix. While the specific location of Gimlet Labs hasn't been disclosed, their innovative technology promises to revolutionize how AI workloads are managed across diverse hardware.