Model Performance

Twitter

Facebook

Copy Link

Key Facts

Key PlatformsLMSYS Arena, Open LLM Leaderboard

Key BenchmarksMMLU, HumanEval, GSM8K, ARC

Top News

Economists Shift Stance on AI and Jobs as New Data Shows Early Signs of Workforce Disruption

After years of skepticism, economists are drawing stronger connections between AI and jobs as Federal Reserve data reveals significant employment slowdowns among software developers. New research from MIT suggests AI's impact on the labor market will unfold gradually rather than catastrophically, but policymakers remain unprepared for the transformation ahead.

Business8 sources

Economists Shift Stance on AI and Jobs as New Data Shows Early Signs of Workforce Disruption

Business8 sources

HubSpot shifts AI agents to outcome-based pricing, charging only for completed results

HubSpot is changing how customers pay for AI by introducing outcome-based pricing for its Breeze Customer Agent and Breeze Prospecting Agent starting April 14. Instead of charging per interaction, customers now pay only when agents successfully complete tasks like resolving support issues or qualifying leads. The move addresses growing pressure on businesses to justify AI spend and prove ROI in an industry where 75% of vendors lack a systematic pricing approach.

Business2 sources

HubSpot shifts AI agents to outcome-based pricing, charging only for completed results

Business2 sources

Microsoft releases three in-house AI models in direct challenge to OpenAI partnership

Microsoft unveiled three foundational AI models—MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2—marking its first major independent release since renegotiating its OpenAI partnership. The models handle speech-to-text transcription, voice generation, and image creation, positioning Microsoft as a direct competitor to its $13 billion investment partner while expanding its proprietary capabilities in the crowded AI market.

Technology16 sources

Microsoft releases three in-house AI models in direct challenge to OpenAI partnership

Technology16 sources

OpenAI's Spud model emerges as major step toward AGI with unified next-gen capabilities

OpenAI is developing an advanced AI model codenamed Spud, representing two years of focused research toward Artificial General Intelligence. The new flagship AI model promises enhanced reasoning and better user intent comprehension, potentially replacing current specialized systems with a unified base model for ChatGPT and enterprise tools.

Technology2 sources

OpenAI's Spud model emerges as major step toward AGI with unified next-gen capabilities

Technology2 sources

Oumi launches platform to automate custom AI models in hours instead of months

Seattle-based AI startup Oumi has launched a commercial platform that automates the development of custom AI models, reducing build time from weeks to hours. Founded by former Google and Microsoft engineers, the company raised $10 million in seed funding and has already gained nearly 9,000 GitHub stars for its open-source toolkit.

Technology2 sources

Oumi launches platform to automate custom AI models in hours instead of months

Technology2 sources

Microsoft Copilot now uses GPT and Claude AI models together to fact-check research outputs

Microsoft unveiled a multi-model approach for its Copilot Researcher tool that leverages both OpenAI's GPT and Anthropic's Claude simultaneously. The new Critique feature has GPT draft responses while Claude reviews for accuracy, achieving a 13.8% improvement on industry benchmarks. The company also rolled out Copilot Cowork to early-access customers as competition intensifies in enterprise AI.

Technology12 sources

Microsoft Copilot now uses GPT and Claude AI models together to fact-check research outputs

Technology12 sources

DeepSeek suffers 7-hour outage, its longest service disruption since R1 and V3 models launched

China's DeepSeek AI chatbot went offline for over 7 hours on March 30, marking the longest outage since its viral debut in early 2025. The disruption affected the platform's web interface, which had maintained near-perfect uptime for over a year. The incident raises questions about platform reliability as the company prepares to unveil its anticipated V4 model.

Technology4 sources

DeepSeek suffers 7-hour outage, its longest service disruption since R1 and V3 models launched

Technology4 sources

AI learns to identify childhood cancer survivors who need extra support through patient conversations

Artificial intelligence could help physicians determine which childhood cancer survivors need additional support by analyzing patient conversations. Researchers at St. Jude Children's Research Hospital found that large language models using complex prompting strategies can detect symptom severity in interviews with survivors and caregivers, potentially transforming how physicians use conversational data to improve survivorship care.

Health2 sources

AI learns to identify childhood cancer survivors who need extra support through patient conversations

Health2 sources

Google launches Gemini 3.1 Flash Live with faster responses and natural-sounding AI voices

Google unveiled Gemini 3.1 Flash Live, its highest-quality audio and voice AI model designed for real-time conversations. The update brings faster responses, more natural cadence, and a doubled context window to Gemini Live and Search Live, which now expands to over 200 countries. All outputs include SynthID watermarks to identify AI-generated speech.

Technology8 sources

Google launches Gemini 3.1 Flash Live with faster responses and natural-sounding AI voices

Technology8 sources

Cohere launches Transcribe, an open-source ASR model that tops accuracy benchmarks

Enterprise AI company Cohere has released Transcribe, its first automatic speech recognition model designed for transcription tasks like meeting notes and speech analysis. The lightweight open-source voice model achieves a 5.42% word error rate, ranking first on Hugging Face's Open ASR Leaderboard while supporting 14 languages and processing 525 minutes of audio per minute.

Technology2 sources

Cohere launches Transcribe, an open-source ASR model that tops accuracy benchmarks

Technology2 sources

Deccan AI raises $25 million to build high-accuracy AI systems for enterprises and frontier labs

Deccan AI has secured $25 million in Series A funding led by A91 Partners to expand its AI model evaluation and enterprise automation services. The startup, founded in October 2024, leverages a network of over 1 million contributors primarily based in India to provide post-training data and evaluation work for frontier AI labs including Google DeepMind and Snowflake.

Startups2 sources

Deccan AI raises $25 million to build high-accuracy AI systems for enterprises and frontier labs

Startups2 sources

Ai2 releases open-source web agent MolmoWeb to challenge OpenAI, Google, and Anthropic

The Allen Institute for AI unveiled MolmoWeb, an open-source web agent built on the Molmo 2 multimodal model that navigates browsers using screenshots rather than code. Available in 4B and 8B parameter versions, the 8B model outperformed GPT-4o on key web navigation tasks, offering developers and researchers a community-driven alternative to closed systems from major tech companies.

Technology2 sources

Ai2 releases open-source web agent MolmoWeb to challenge OpenAI, Google, and Anthropic

Technology2 sources

Telling AI Models They're Experts Actually Makes Them Worse at Knowledge Tasks

University of Southern California researchers found that popular expert persona prompts harm AI model accuracy on factual tasks. While instructing AI to act as an expert helps with writing and safety, it degrades performance on math and coding by shifting models into instruction-following over factual recall mode. A new technique called PRISM aims to solve this tradeoff.

Science3 sources

Telling AI Models They're Experts Actually Makes Them Worse at Knowledge Tasks

Science3 sources

Microsoft's MAI-Image-2 cracks top three on Arena.ai leaderboard behind Google and OpenAI

Microsoft unveiled MAI-Image-2, its second-generation AI image generation model, which debuted at #3 on Arena.ai's text-to-image leaderboard. The model excels at photorealism, accurate text rendering, and complex scene generation. It's now rolling out across Microsoft Copilot, Bing Image Creator, and MAI Playground, marking Microsoft's shift from relying on OpenAI's models to building competitive in-house technology.

Technology5 sources

Microsoft's MAI-Image-2 cracks top three on Arena.ai leaderboard behind Google and OpenAI

Technology5 sources

Cursor admits Composer 2 was built on Chinese AI model, exposing Western open-source gaps

AI coding startup Cursor launched Composer 2 as frontier-level coding intelligence but failed to disclose it was built on Moonshot AI's Kimi 2.5 model. A developer exposed the connection within hours by intercepting API traffic. The incident highlights why Western companies increasingly turn to Chinese open models when competing with larger AI firms.

Technology10 sources

Cursor admits Composer 2 was built on Chinese AI model, exposing Western open-source gaps

Technology10 sources

Prev 1…101112…24 Next