Model Performance

Twitter

Facebook

Copy Link

Key Facts

Key PlatformsLMSYS Arena, Open LLM Leaderboard

Key BenchmarksMMLU, HumanEval, GSM8K, ARC

Top News

OpenAI rolls out GPT-5.3 Instant with fewer refusals and less preachy tone across ChatGPT

OpenAI released GPT-5.3 Instant, addressing user complaints about overly cautious responses and unnecessary refusals. The update reduces hallucinations by 26.8% with web search and delivers more direct answers without moralizing preambles. While the model shows improvements in conversational flow, CEO Sam Altman acknowledges ongoing challenges with model personality across the GPT-5 family.

Technology9 sources

OpenAI rolls out GPT-5.3 Instant with fewer refusals and less preachy tone across ChatGPT

Technology9 sources

Google launches Gemini 3.1 Flash Lite with 2.5x speed boost and adjustable thinking levels

Google unveiled Gemini 3.1 Flash Lite, its fastest and most cost-efficient AI model designed for high-volume developer workloads. Priced at $0.25 per 1M input tokens, the model delivers 2.5x faster time to first token than its predecessor while introducing adjustable Thinking Levels that let developers balance speed with reasoning depth for real-time applications.

Technology11 sources

Google launches Gemini 3.1 Flash Lite with 2.5x speed boost and adjustable thinking levels

Technology11 sources

Humanity's Last Exam reveals the gap between AI and human intelligence despite rapid progress

Researchers created Humanity's Last Exam, a PhD-level AI benchmark with 2,500 questions designed to assess the limits of AI reasoning. Google's Gemini 3 Deep Think achieved the highest score at 48.4%, while human experts score around 90%. Despite progress, experts emphasize this doesn't signal the arrival of Artificial General Intelligence.

Science2 sources

Humanity's Last Exam reveals the gap between AI and human intelligence despite rapid progress

Science2 sources

Pantera, Franklin Templeton Back Sentient Arena to Test AI Agents on Enterprise Workflows

Open-source AI lab Sentient launched Arena, a production-grade platform to stress-test AI agents on complex enterprise tasks. Pantera Capital, Franklin Templeton's digital assets unit, and Founders Fund joined the first cohort, signaling institutional interest in evaluating AI agent reliability before deployment into real-world workflows.

Technology2 sources

Pantera, Franklin Templeton Back Sentient Arena to Test AI Agents on Enterprise Workflows

Technology2 sources

Harvard Study Shows AI Can Predict 71% of Fund Manager Trades, Raising Questions About Fees

A Harvard-led study reveals that AI can predict 71% of mutual fund trading decisions, suggesting much of active management follows detectable patterns. The research, analyzing data from 1990 to 2023, found that the unpredictable 29% of trades is where genuine alpha lives. The findings challenge the justification for high active-management fees and highlight how automation could reshape the $54 trillion asset management industry.

Science3 sources

Harvard Study Shows AI Can Predict 71% of Fund Manager Trades, Raising Questions About Fees

Science3 sources

ChatGPT Health fails to recognize half of medical emergencies in first independent safety test

OpenAI's ChatGPT Health missed over half of medical emergencies in a Nature Medicine study, directing patients to routine appointments instead of emergency rooms. With 40 million daily users seeking health guidance, the AI tool also showed alarming inconsistencies in suicide-crisis safeguards, triggering alerts for low-risk cases while failing to respond when users described specific self-harm plans.

Health9 sources

ChatGPT Health fails to recognize half of medical emergencies in first independent safety test

Health9 sources

AI model predicts bone removal in cochlear implant surgery to enhance surgical planning

Researchers from Vanderbilt University and partners developed an AI model that predicts bone removal during cochlear implant surgery. The system analyzes CT scans to forecast mastoidectomy shape, achieving a 0.72 Dice score across 751 cases. This medical imaging technology could support robotic surgical tools and improve patient outcomes.

Science2 sources

AI model predicts bone removal in cochlear implant surgery to enhance surgical planning

Science2 sources

Elon Musk delays Grok update after chatbot fails Baldur's Gate 3 knowledge test

Elon Musk reportedly delayed a Grok model release for several days because the xAI chatbot couldn't properly answer detailed questions about Baldur's Gate 3. High-level engineers were pulled from other projects to improve Grok's gaming knowledge, while separate teams worked in war rooms to teach Grok to play League of Legends.

Technology4 sources

Elon Musk delays Grok update after chatbot fails Baldur's Gate 3 knowledge test

Technology4 sources

Google Gemini 3.1 Pro doubles reasoning score, beats rivals in key AI benchmarks

Google released Gemini 3.1 Pro with dramatically improved problem-solving and reasoning capabilities, more than doubling its predecessor's performance on abstract reasoning tests. The newest and more powerful large language model scored 77.1% on ARC-AGI-2 and achieved a record 44.4% on Humanity's Last Exam, outperforming OpenAI and Anthropic in most benchmarks as accelerating competition among major tech companies intensifies.

Technology27 sources

Google Gemini 3.1 Pro doubles reasoning score, beats rivals in key AI benchmarks

Technology27 sources

MIT Researchers Expose Hidden Biases and Personalities in Large Language Models

Researchers from MIT and UC San Diego have developed a method to identify and manipulate over 500 hidden concepts within large language models, including biases, personalities, and moods. The technique uses a Recursive Feature Machine algorithm to steer AI responses, improving performance while also exposing potential vulnerabilities like jailbreaking and hallucinations.

Science3 sources

MIT Researchers Expose Hidden Biases and Personalities in Large Language Models

Science3 sources

OpenAI and Paradigm Launch EVMbench to Test AI Agents on Ethereum Smart Contract Security

OpenAI and crypto investment firm Paradigm unveiled EVMbench, a benchmark tool designed to evaluate how AI agents detect, patch, and exploit vulnerabilities in Ethereum smart contracts. Drawing on 120 real flaws from 40 audits, the tool tests models like GPT-5.3-Codex, which scored 72.2% in exploit mode, as billions in crypto assets remain at risk.

Technology3 sources

OpenAI and Paradigm Launch EVMbench to Test AI Agents on Ethereum Smart Contract Security

Technology3 sources

Tempus AI stock rises 3.5% after unveiling HRD-RNA algorithm for cancer detection

Tempus AI shares climbed Wednesday following the launch of its new HRD-RNA algorithm, an AI-driven tool designed to detect homologous recombination deficiency in tumors. The algorithm analyzes gene expression to identify patients who may benefit from platinum-based chemotherapy or PARP inhibitors, offering a more dynamic assessment than traditional genomic methods.

Health2 sources

Tempus AI stock rises 3.5% after unveiling HRD-RNA algorithm for cancer detection

Health2 sources

Sarvam AI Launches Indus Chat App as Indian Startup Takes on ChatGPT and Gemini

Bengaluru-based Sarvam AI has launched its Indus chat app, powered by a 105-billion-parameter model built specifically for India's 22 official languages. The startup trained its AI models for Indian market with just 40 researchers and 4,000 GPUs over six months, positioning itself as a frugal alternative to global rivals like ChatGPT and Gemini while advancing India's sovereign AI ambitions.

Technology9 sources

Sarvam AI Launches Indus Chat App as Indian Startup Takes on ChatGPT and Gemini

Technology9 sources

Anthropic releases Claude Sonnet 4.6 with human-level computer use and coding improvements

Anthropic launched Claude Sonnet 4.6, bringing frontier-level AI capabilities to free and Pro users. The update delivers improved coding skills that rival its premium Opus model, human-level performance in computer use tasks, and a 1 million token context window. Developers prefer the new model 70% of the time over its predecessor, with major gains in security against prompt injection attacks.

Technology18 sources

Anthropic releases Claude Sonnet 4.6 with human-level computer use and coding improvements

Technology18 sources

Tesla's Full Self-Driving tries to drive owner into lake, raising fresh safety concerns

A Tesla owner captured his vehicle attempting to drive straight into a lake while using Full Self-Driving version 14.2.2.4. Daniel Milligan's viral video, which garnered over 1.2 million views, shows the car accelerating down a boat ramp at night before he intervened. The repeatable incident adds to mounting regulatory pressure as NHTSA investigates 2.88 million Tesla vehicles.

Technology3 sources

Tesla's Full Self-Driving tries to drive owner into lake, raising fresh safety concerns

Technology3 sources

Prev 1…121314…24 Next