Research
OpenAI Finds 30 Percent of Popular AI Coding Test Broken
OpenAI has pulled its endorsement of SWE-Bench Pro after finding 30 percent of its tasks flawed, impacting AI performance assessments.
LATEST NEWS
-
AMD Introduces FlyDSL for High-Performance GPU Kernel Development
AMD released FlyDSL, a Python DSL for GPU kernel development, enabling performance matching hand-tuned C++ kernels with reduced complexity. The tool is tested on AMD Instinct MI355X GPU using ROCm 7.2.2.
-
Mistral Studio Offers Governance for AI Prompts and Skills
Mistral announced on July 9, 2026, that Studio provides a centralized system of record for managing AI prompts and skills, ensuring traceability and compliance.
-
Meta Launches Muse Image and Muse Video Models
Meta introduced Muse Image and Muse Video, its first media generation models, with Muse Image ranked second in text-to-image benchmarks as of July 5, 2026.
-
Anthropic Introduces Claude Reflection Feature
Anthropic launched a new reflection tool for Claude users, offering insights into AI integration in daily life, available in beta for Free, Pro, and Max users.
-
Databricks Uses GLM 5.2 as Default Coding Engine
Databricks found GLM 5.2 performs on par with Opus 4.8 but at a lower cost, leading to its adoption as the default coding model.
-
Nandan Nilekani Steps Down as Fundamentum Launches Third $200M Fund
Nandan Nilekani, co-founder of Infosys, is stepping down from his general partner role at Fundamentum as the firm launches its third $200 million fund.
-
Ollama Raises $65M in Series B, Grows to Nearly 9M Users
Ollama, the open-source AI developer tool, raised $65 million in Series B funding, growing its user base to nearly 9 million.
-
OpenAI AI Dominates AtCoder Contest, Solving All Five Problems
OpenAI's AI system outperformed all human competitors at the 2026 AtCoder World Tour Finals, solving all five Algorithm Division problems, including two rated exceptionally difficult.
-
Distillation Techniques in 2026: How Major Models Use the Method
Distillation is central to 2026's frontier models, with Gemma 4 and DeepSeek-V4 using it to compress large models and merge domain-specific expertise.
-
Lovable Talks to Double Valuation to $13.2 Billion
Lovable, a Swedish vibe-coding startup, is reportedly in talks to raise $300 million at a $13.2 billion valuation, doubling its previous $6.6 billion valuation from December.
-
Prime Intellect Lets Users Build Self-Improving AI Models
A startup called Prime Intellect is letting users build self-improving AI models, creating a more distributed approach to AI development. The company raised $15 million in funding recently.