#Benchmarks

UniPat AI launches the most stringent wet experiment prediction benchmark SPU, with cutting-edge models performing poorly

UniPat AI has released a new benchmark SPU, including 116 tasks and 4207 wet experiments, covering fields such as chemistry and materials science, requiring models to conduct autonomous research and predict experimental results. Evaluations show that Claude Opus 5.5 scored an average of 0.486, with only 34%-52% of tasks exceeding random levels, indicating that AI still has significant room for improvement in wet experiment prediction. The SPU evaluation emphasizes the model's autonomous research process and prediction performance, providing a new standard for evaluating AI research capabilities.

Former Anthropic researcher Coxon: Anthropic can now retrain the Mythos model using 'a small portion of the original computing resources'

Former Anthropic and OpenAI researcher Jacob Coxon said the IPO roadshow slides for Anthropic were the main reason for his departure, as they showed that after a major leap in capabilities for its Mythos model in the spring, it could be retrained using 'a small portion of the original computing resources,' indicating that algorithm efficiency was accelerating.

OpenAI has reassigned robotics researchers to other teams as general models may make dedicated robotics AI unnecessary

OpenAI has reorganized its robotics team again, assigning researchers to other departments instead of dissolving the team. Robotics researchers will join the main research team led by Chief Research Officer Mark Chen, while the application robotics hardware team will join Greg Brockman's application department.

SO-Arena evaluates the capability spectrum and reliability of a model

SO-Arena evaluates system one model through 11 domains and 5 capability families, finding differences in scalability and reasoning gain, and high overall accuracy does not guarantee reliable decision-making. This benchmark assesses the capability spectrum through reliable sources and manual verification, domain-specific difficulty standards, and iterative execution of actions in a sandbox; it evaluates decision reliability through decision stability under option set changes and answer identification when there are no valid options. 21 model evaluations show differences in scalability and reasoning gain across domains, and the best system one model ranks only seventh in reliability.

NVIDIA announces CEO Huang Renxun's resignation, CFO Colette Kress takes over as interim CEO

NVIDIA announces CEO Huang Renxun's resignation, CFO Colette Kress takes over as interim CEO; President Trump suggests Cook for the position; Tencent denies 50% layoffs; Godiva's operational partner fined 800,000 for promoting incorrect views on marriage and relationships.

Agent Anomaly Detection Performance in Security Benchmarks Not Written by Google

Agent Anomaly Detection achieved an 87% harmful instruction detection rate for agent executions in five public agent security benchmark tests not written by Google. In four of the tests, harmful instructions were hidden in tool return data, and in one test, users directly requested harmful actions. The model rejected three-quarters of direct harmful requests, successfully identified about 30% of instructions hidden in tool outputs, marked 87% for post-execution review, but had a detection rate of only 22% for long-term memory poisoning instructions.

The study found that AI chatbots recommend fake retail websites.

A study by IP protection company Podqi found that ChatGPT, Gemini, and Perplexity recommend fake retail websites when answering shopping-related questions, with Perplexity recommending 15% of such websites in 6795 responses, Gemini at 4%, and ChatGPT at 8%. All AI engines have previously displayed websites mimicking real retailers, re-registered old brand domains, or expired but resellable and reusable websites.

Baidu's Robin Li launches the 'Word Element Exchange': claims neutrality and credibility, rating model providers

Baidu's Li Yannhong launched the large model service platform 'Ci Yuan Suo' at the Baidu World Conference 2026, claiming to solve the problems of difficult model selection and varying quality. Ci Yuan Suo ensures model reliability through comprehensive entry tests, continuous quality monitoring, and a multi-dimensional scoring system, and provides a credit record for model providers. One month after launch, the platform has gathered 100 supply resources, with daily average Token calls reaching 700 billion.

LG T6 Transparent OLED and W7 Wallpaper OLED both win CES Innovation Award

The LG SIGNATURE OLED T6 won the 'Best Innovation Award' in the Art and Technology category, featuring a 77" 4K transparent OLED panel and supporting 60GHz millimeter-wave Zero Connect; the new wireless picture frame TV LG OLED evo AI W7 won the Innovation Award in the Video Display category, with a thickness of 15.7mm and supporting a 165Hz refresh rate.

Genie with Genie Ontology performs excellently in enterprise data proxy benchmark tests

Development progress of IBM Quantum Starling

IBM Quantum Starling will be delivered in 2029, capable of performing 100 million hard operations on 200 qubits. IBM has validated the c-coupler technology, achieving dense non-local qubit connectivity, and has improved the qLDPC code to require only five qubit connections instead of six.

Reddit launches AI summary system for product reviews, causing user concerns

Reddit tests its Reviews feature, which uses AI to summarize user reviews to analyze the pros and cons of products, but users are concerned that it may harm genuine communication and promote soft advertising. Moderators believe it will blur the line between genuine opinions and advertising. Reddit claims that embedded links are quick access, not paid advertising spots, and brands cannot pay to get display qualifications.

Google updates Android AI development benchmark list: Kimi K3.1 Pro / DeepSeek V4.1 Flash scores revealed

Google has announced the Android AI Development Benchmark (Android Bench) score, where Kimi K3.1 Pro passed 24.0% of 30 tasks using the kimi-code intelligent agent.

Google, in collaboration with Artificial Analysis, has launched the MKA benchmark: Gemini 4 Carbon multimodal knowledge work test ranked second, with task costs being only one-sixth of the first place.

Google, in collaboration with Artificial Analysis, has launched the MKA benchmark test to evaluate the ability of AI agents to handle multimodal knowledge work. Gemini 4 Carbon ranked second in the test, with task costs being one-sixth of the first-place cost.

Claude Fable 5.2 leads multimodal knowledge work in Multimodal Knowledge Agent (MKA) evaluation

Claude Fable 5.2 (max) leads in the Multimodal Knowledge Agent (MKA) evaluation with a 57% Climb WSR score, followed by Gemini 4 Carbon (high) at 54% and Claude Opus 5.5 (max) at 53%.

Google DeepMind's Gemini Robotics 2.5 achieves a success rate of 96% in Mercedes-Benz assembly line trials

OpenRouter Test: Claude Opus 5.5 is the best OCR model for scanning documents

Claude Opus 5.5 identified 262 out of 264 key values in OpenRouter testing, leading with a price of $16.42 per thousand pages, Mistral OCR ranked second with 258 key values and a cost of $2.02 per thousand pages, and Ling 3.0 Flash VL performed best with 254 key values and a cost of $0.04 per thousand pages due to disabled reasoning.

Fable 5.2 Initial Tests: Significant Ability Boost but 20% Price Drop, Users Create Stunning Effects in Games, Videos, 3D Modeling

Seedream 5.5 released: Domestic AI models perform excellently in continuous image editing tasks

Seedream 5.5 outperformed GPT-Image 2.6 in high-pressure tasks of continuous image editing for 100 rounds, showing greater stability. The model enhances realism, supports small text and multi-language rendering, transparent image generation and editing, as well as layer splitting and text secondary editing functions. Additionally, Seedream 5.5's API price has dropped to 0.28 yuan per image, with improved speed, maintaining the 'input image + output image' two-stage billing structure, and no text token fees.

How is Inference Performance Index (IPI) Measured for Inference Capability

Nvidia GB300 offers a 25-fold improvement in capability compared to H100 for intelligent agent inference tasks, far exceeding the 2.5-fold increase in 8-bit FLOP/s. The IPI index measures inference capability by comparing the number of requests a chip can handle simultaneously under various output speed requirements. GB300 is set as the baseline value of 1, and chips scoring 1 can serve twice as many users under the same conditions. IPI version 1 combines results from four representative models, using GB300 as the reference benchmark.

Introducing the Artificial Analysis Decision Index v2

Artificial Analysis Decision Index v2 introduces a new metric rewarding model correctness and calibration confidence, with a 45% benchmark based on a private dataset; Drex Max v1.1 (Nace AI) leads with 73 points, GPT-6 Luna (Max) ranks first among language models with 76 points, but with significantly higher costs and inference time.