Zhiyuan GO-3 Nano Preview tops RoboDojo evaluation with an average score of 61.90 and a success rate of 56.42%, surpassing GPT-6 Astra, and ranking first in fine manipulation, long-range execution, and open semantic understanding across three sub-dimensions.
The DecoderOpenAI releases GPT-6.1 Terra, pushing the cost-performance boundary to new heights, API prices halved compared to GPT-5.6 Terra to 1 dollar per million input tokens, 5 dollars per output token
Artificial Analysis ArticlesFeaturedSame story · 2 outlets8585OpenAI released GPT-6.1 Terra, the first Terra-level model after GPT-5.6, with input/output pricing at $1/5 per million tokens, lower than the $2/12 per million tokens of GPT-5.6 Terra.
How the accounts differ:Artificial Analysis Articles states the input/output pricing is $1/5 per million tokens for GPT-6.1 Terra, while The Decoder mentions it is $2/12 for GPT-5.6 Terra.
Why it matters: The original text used the scores of the Intelligence Index and Coding Agent Index, along with the cost per task, to quantify the change in cost-effectiveness of GPT-6.1 Terra compared to GPT-5.6 Terra.
OpenAI Deployment Safety5858GPT-6.1 Terra uses the same data and training methods as GPT-6 Astra. It has been evaluated as highly capable in the fields of biology and chemistry, as Multimodal Troubleshooting Virology has crossed the instruction threshold.
Artificial Analysis Articles6565Claude Opus 5.5 leads in five metrics on Code2Video-AA, with cost differences reaching 300 times per task, and all model video rendering rates at least 96%.
Anthropic has launched Claude Fable 5.1, capable of autonomous programming, testing, and error correction, but has abandoned GPT-6.1 Astra due to its boundary-crossing behavior. AI is moving from 'teaching by word and example' to 'generalizing by analogy', exploring autonomy through Human-in-the-Loop (HIL) and Recursive Self-Improvement (RSI). Claude leads about 26% of internal R&D work. Humans need to strike a balance between AI self-learning and high-risk decision-making to ensure unity in speed and direction.
Research found that text generated by AI models uses language features more frequently associated with understanding difficulty, such as long syntactic dependencies, low-frequency n-grams, and the use of inanimate subjects with active verbs, compared to human writing. Compared to human writing, 9 cutting-edge and near-cutting-edge models use more low-frequency n-grams, most of which also use longer dependencies, more stacked nouns, and more ellipsis of object clauses, with significant differences among models. The study suggests that AI text is optimized for goals such as mathematics, code, and reasoning, leading to information density prioritizing effective communication, and this hypothesis is verified by fine-tuning language models.
OpenAI and Anthropic have reduced the prices of LLMs by up to 80% within a few months. OpenAI has reduced the price of GPT-5.6 Luna by 80% to $0.20 / $1.20 / $0.02, and the input and output prices of GPT-6 Sol and Luna have been reduced by 50% to $2 / $10 / $0.20 and $0.10 / $0.50 / $0.01 respectively.
Toyota plans to deploy 400,000 robots, with an annual investment of $6.4 billion for factory automation, forming a dedicated team led by the chief production officer; its spin-off Walden Robotics has completed $300 million in funding, valuing at $1.1 billion, and its General Robotics has been deployed in North American factories. Toyota has showcased an AI-powered humanoid robot Cue7 and developed large behavioral models, aiming to become the world's largest humanoid robot company, with an estimated market share of 3%-8% by 2035.
Browser Use3838The benchmarking methods for AI agents on websites like WebArena, WebVoyager, and Mind2Web differ. WebArena and similar sandbox environments provide reproducible scores, while real website environments like WebVoyager are scored by humans or LLMs. Offline datasets like Mind2Web compare prediction steps with human behavior. Browser Use has released BU Bench V2.1, which uses GPT-5.6 Luna for scoring in real website environments and includes model cost and time data.
InterconnectsSame story · 3 outlets6565Kimi K3.1 Pro release has sparked controversy, its benchmark tests surpassing GPT-6 Astra on Artificial Analysis but slightly underperforming on Terminal-Bench Science and ProgramBench. The model release is accompanied by suspected paid promotion, and its pre-training effectiveness is criticized as not meeting expectations.
How the accounts differ:Artificial Analysis gives the score 55 and ranks fourth; MarkTechPost mentions the total parameters of 2.8T and the MoE model with 104B active parameters.
The Decoder6565Anthropic founder Clark, based on TasteVal benchmark test results, has moved the timeline for AI self-improvement (RSI) from 2028 to 2027. TasteVal measures the computational efficiency of AI models in scientific research tasks. Anthropic's Opus 5.5 model achieves expert level with a 2.3 computational multiplier. Since December 2025, the research taste of top AI models has doubled every three months.
The DecoderFeatured7272Researchers from Nanjing University and institutions like SenseTime have released OpenMobile-2, which includes two AI models based on Qwen3.5 (9B) and Qwen3.6 (27B), and has publicly disclosed the training environment, data, and reproduction instructions.
Why it matters: Research fixed the training data size while only increasing the number of applications, observing an improvement in success rates for unseen applications, providing a reproducible path for data construction of mobile agents.
Latent SpaceSame story · 3 outlets6767How the accounts differ:Artificial Analysis gives the score 55 and ranks fourth; MarkTechPost mentions the total parameters of 2.8T and the MoE model with 104B active parameters.
Big TechnologySundar Pichai believes the tech industry is far from reaching the end of recursive self-improvement (RSI), although AI models are achieving automated improvements with the assistance of AI researchers; he claims the entire tech industry is facing a shortage of computing resources due to a decision two years ago, Google Cloud Platform is growing rapidly but needs to sacrifice short-term profits for cutting-edge AI research, and believes moderate constraints can inspire AI innovative thinking.
Simon WillisonFeatured8282Kimi K3.1 Pro goes live with Kimi API today, weights not yet open, Moonshot promises to release complete model weights by November 4th, 2026. Pricing is $2.00/100 million tokens for input, $10/100 million tokens for output, the same as GPT-6.1 Sol, half of Claude Opus 5.5, also provides a Highspeed version that is 2-3 times faster and doubles the price.
The DecoderFeatured8585Why it matters: The article uses third-party indices and multiple benchmarks to present the specific extent to which the gap between open-source weight models and closed-source cutting-edge models has narrowed.
The DecoderAnthropic claims its unreleased model is superior in both capabilities and alignment to Mythos 5.1, with the model able to stop attacks on its own 67% of the time in a cybersecurity test replicating the UK AI Safety Institute (AISI), while Mythos 5.1 only managed 10%
Hugging Face Daily Papers6565Code2Video-Compass benchmark includes 2,104 tasks, covering code video generation from text, image, and video inputs, evaluating 15 MLLMs in T2V, I2V, and V2V scenarios, and finding that top models still face challenges in generating high-quality videos for complex dynamic and state-dependent tasks.
The throughput of domestic large models reached 500 trillion tokens per second in June 2026, an increase of nearly 4 times compared to December 2025, but the average TPS in August was 78.23 tokens per second, still lower than the foreign 112.82 tokens per second.
TechCrunch · AIMicrosoft's AI diagnostic system MAI-DxO achieves 83% accuracy in simulating complex cases, surpassing general practitioners working alone (36%), by coordinating multiple AI agents to simulate a medical team's operation.
The Verge · AIMicrosoft claims its AI tool has four times the diagnostic accuracy of human doctors in simulated medical scenario tests, with related research published in the journal 'Science'. The tool is called Diagnostic Orchestrator, which collaborates multiple AI agents to reason symptoms and make diagnoses, achieving 83% accuracy in medical report tests and 80% in emergency data tests, superior to human doctors by about 47%.
When 13 AI models were asked about their 'controversial' views on humans, 32 gave the same answer. GPT-6.1 Sol, Grok 4.7, DeepSeek V4 Pro, Kimi K3.
Sierra Blog7575The 𝜏-Entity benchmark speech intelligence agent captures the accuracy of entity information such as names and phone numbers in 200 tasks. GPT Live 1 has a correct rate of 60.3% under normal conditions, but only 34.0% under noise and voice interference; recommended verification strategies can increase success rates by 14-31 percentage points, but call duration increases by 21-28 seconds.
Hugging Face Daily Papers6565StreamEval benchmarked the performance of interactive streaming video generation models on five tasks, finding reliable request execution and sustained state consistency to be major challenges. The benchmark includes 300 scripted test cases, covering character dialogue, character performance, world generation, instruction editing, and reference editing tasks, scored across four evaluation dimensions (task execution, state consistency, visual fidelity, and audiovisual coordination).
O'Reilly Radar7575Anthropic officially announced that future Claude models will include AI text watermarks to comply with the EU AI Act, including Claude Mythos 5.1 and Claude Fable 5.1.
Boundless Dynamics has released and open-sourced the MWA™-Phoenix Subspace World Model Foundation, sweeping the LIBERO-Plus, LIBERO-Pro, RoboCasa GR1, Bench2Dex, RoboTwin 2.0, and RoboDojo real machine rankings.
MarkTechPost5555The Cactus Whistle speech model transcribes seven languages on the CPU with a 16.9 MB file size and returns word timestamps and probabilities. Tests show it outperforms Whisper Base on its test-other dataset, but there is no significant difference from beam search, with the speed advantage shrinking as audio length increases. Additionally, it tested 30-second window audio processing, streaming latency, timestamps, keyword bias, other languages, noise, embedding, and speech-to-tool calling features.
AWS Machine Learning Blog3131Since its launch, Amazon SageMaker JumpStart has integrated over 100 new models from 30 providers, including Qwen 3.8-27B, Cosmos3 series, etc.
OpenAI News4242Terminal X's Eval Agent achieves an 88.6% score using GPT-6.1, based on 54 criteria, able to discover 25% more fund mission deviations than other solutions
The Decoder7575MiniMax released the M3.1 Flash model, scoring 42 on the AI analysis index, an increase of 13 points over the previous generation M3, with a task cost less than half of the previous generation. This model is ranked alongside GLM-5.3-Flash, surpassing Gemini 3.8 Flash (41 points) and second only to Deepseek V4.1 Flash 1017 (45 points).
Big TechnologyGoogle CEO Sundar Pichai said that his latest released Gemini 4 Argon model has outperformed OpenAI and Anthropic's latest public models in coding, mathematics, science, and knowledge work benchmarks, bringing Google AI back to the top tier in the world.
OpenAI Alignment Blog5050OpenAI has introduced the complementary metric g-mean² to improve model monitorability evaluation. g-mean² calculates TPR and TNR through behavioral labels and monitoring predictions, but cannot solve the problems of mixed differences in different model tasks and deployment risks. The new metric includes a paired matching metric for aligned shared instances and a post-control integrated metric simulating monitoring rejection of outputs, to more comprehensively evaluate model deployment risks.
Hugging Face Blog5858FairStress benchmark evaluated LLM performance in demographic fairness across 300 decision scenarios across 29 fields (such as recruitment, healthcare), including 13 million test cases. The study found that the same persuasive prefixes (adversarial prefixes) have varying impacts on model decisions across different demographic groups, and there are differences in how models explain decisions containing demographic information.
One Useful Thing7575Claude Code v2.1.159 released, supporting an 8k context window, available on GitHub; the author generated charts in Claude Projects using Opus 5.5, where AI handled time-consuming research, drawing, layout, image, and citation work, demonstrating exponential growth in AI capabilities.
Artificial Analysis Articles7070AA-PersonalityBench evaluates 30 personality traits, including friendliness, through 900 questions, using LLM judges for blind pairwise judgments, providing Elo scores and 95% confidence intervals; Claude Sonnet 5.5 scores highest in traits like friendliness (Elo 1000), and Gemini 4 Argon (high) performs outstandingly in traits like transparency (Elo 1059).
