#Reasoning

How Huawei’s Kirin 9050 Pro logic folding supports Mate 90 flagship features

Huawei says its mass-produced Kirin 9050 Pro uses logic folding to shorten signal paths through vertically connected circuit layers, supporting performance and efficiency gains in the Mate 90 Pro Max. The article explains how dense hybrid bonding enables this layout and how memory coordination allows Kirin 9050 chips to run a 30B MoE model with less than 2.5GB of additional memory usage. Huawei also says Kirin 2027 chips are in testing, with results meeting expectations so far.

vLLM Adds Adaptive Verification for Draft Models Without a Confidence Head

Up until a month ago, adaptive verification in vLLM needed a drafter that shipped with a trained confidence head. Now it works with the draft models you already serve: MTP, EAGLE, DFlash, and DSpark checkpoints without a head. No retraining. The signal was already in the draft

Gemini 4 Argon Is Unlocking Scientific Breakthroughs

Gemini 4 Argon is already unlocking new scientific breakthroughs in quantum computing, nuclear fusion, protein design and beyond. 🧵 https://t.co/Shj3Y4whoK

Google releases Deep Review for analyzing complex technical documents

Google Research has publicly released Deep Review, an agentic AI system for technical verification and feedback on research papers and other complex documents, including legal and financial materials. Evolving from the Paper Assistant Tool, which provided optional feedback on more than 16,000 manuscripts, it combines document segmentation, adaptive compute budgets, parallel verification, and global synthesis. Users worldwide receive two complimentary reviews per day, with higher limits and API integration available through a Gemini API key; uploaded documents are never used to train, fine-tune, or improve Google models.

Google evaluates Earth AI geospatial agents and expands experimental access

Google released a technical report evaluating its experimental Geospatial Reasoning agent across four benchmarks and is expanding early access through Ask Google Earth and an Agent-to-Agent API. Evaluations using Gemini 3.8 Flash and Gemini 3.1 Pro found that specialized Earth AI tools improved spatial reasoning, while upfront planning helped with open-ended workflows. Google is releasing development splits for GeoKTools and GeoExpertQueries, but cautions that GeoExpertQueries evaluates adherence to expert workflows rather than final-answer correctness, making agent outputs drafts for expert review.

Alexandr Wang Calls Muse Wonder’s Interaction Style “Active Focus”

Muse Wonder: “It’s genuinely one of the first times using a model has felt like telling an employee what to do rather than babysitting a dumbass pet that I have to guide through every step” We call that ACTIVE FOCUS ™️ https://twitter.com/harris_p10/status/2110387905550024815

Alexandr Wang Says Muse Wonder Is Fast and Token-Efficient

@merge_api says Muse Wonder completes an average task in 6.8 minutes, compared with 13.2 minutes for Claude Opus 5.5 and 15.3 minutes for Claude Sonnet 5.5. The quoted post says the scores are 52.4, 57.6, and 56.0, respectively. Alexandr Wang calls Muse Wonder “super fast and token efficient” and says, “give it a go!” https://twitter.com/merge_api/status/2110385332046762163

François Chollet Says Autonomous Robotics Requires Causality and Discrete Reasoning

You can't solve real-world physical tasks through autoregressive sampling alone. The physical world is entirely unforgiving to hallucination or approximate pattern matching. If we want actual autonomous robotics we have to engineer causality and discrete reasoning back into the https://t.co/UJtL2t0tQ2

vLLM Says Practical Speculative Decoding Benefits the Inference Community

Making speculative decoding easier to put into practice benefits the whole inference community. Check out @huggingface's livestream with @mervenoyann, @MaartenGr and @_EldarKurtic for a practical look at speeding up local LLMs! https://twitter.com/huggingface/status/2110350304613114226

Learn about how to get multipliers in local LLM speed-ups using speculative decoding in this livestream with @mervenoyann, @MaartenGr and @_EldarKurtic https://www.youtube.com/watch?v=H4azmWcqO5g

Learn about how to get multipliers in local LLM speed-ups using speculative decoding in this livestream with @mervenoyann, @MaartenGr and @_EldarKurtic https://www.youtube.com/watch?v=H4azmWcqO5g

AA-AgentPerf adds DeepSeek V4.1 Flash benchmarks across NVIDIA and AMD hardware

Artificial Analysis added DeepSeek V4.1 Flash to AA-AgentPerf, its benchmark replaying real agentic coding sessions with prompt caching and tool-call delays. At 100 tokens/s per user, GB300 leads with approximately 261M tokens/s per MW, followed by B300 at approximately 220M and MI355X at approximately 128M. Compared with DeepSeek V4 Pro 0813, V4.1 Flash scores 39 versus 36 on the Artificial Analysis Intelligence Index and delivers approximately 1.3x to 6.2x the throughput per MW at that user output speed, depending on hardware.

Jiayuan Mao explains single-demonstration robot learning and 83% real-world generalization at IROS 2026

Reporting on Jiayuan Mao’s IROS 2026 presentation, Leiphone describes research from UPenn, MIT, and Stanford on robot learning and compositional generalization. RAPID combines a coding agent with physics validation to learn complex non-grasping tasks from one human demonstration. In separate video-action research, the team introduced Temporal Ratio to measure attention to future versus current frames; adaptive guidance based on this metric raised average success on unseen compositional tasks with a real dual-arm robot from 71.8% to 83% without retraining.

Qwen celebrates @lightseekorg’s TokenSpeed results for Qwen3.8-Flash-Next

@lightseekorg reports 685 output tokens/s per user for Qwen3.8-Flash-Next. Congrats to @lightseekorg on the impressive TokenSpeed results! 🚀 Great to see Qwen3.8-Flash-Next reaching 685 output tokens/s per user on Nvida GPUs. Excited to see the community unlock more of Qwen’s inference potential!

Industrial Next wants factory robots to switch tasks in minutes, not weeks

Industrial Next presented AI-driven robotic stations at TechCrunch Disrupt 2026 that it says let manufacturers retool production lines in minutes rather than weeks without traditional programming. Its proprietary foundation models connect vision, task context, and motion, with new tasks taught through simulation and demonstration. The startup announced a partnership with BMW Group Technology Office to evaluate dexterous manipulation for manufacturing, and co-founder Lukas Pankau said some stations already perform delicate assembly tasks for an undisclosed smartphone customer.

Nathan Lambert Calls Situational Awareness Eerie but Not Remarkable

I feel like this level of situational awareness isn't that remarkable, but the interaction still reads as very eerie. I bet pretty old models with "be safe, you are an AI, we are watching" would do stuff like this too.

Dyna Robotics’ Jason Ma Explains Dyna-2 and Hour-Long Physical Agent Workflows

Dyna Robotics co-founder and Chief Scientist Jason Ma explains how Dyna-2 combines future video prediction with robot actions and how agentic harnesses support hour-long workflows. He says the model was trained on over one million hours of human video, with some tasks requiring as little as 45 minutes of demonstrations. The interview covers napkin-folding deployments at Din Tai Fung, on-device inference, and hardware durability and missing sensory data as constraints on broader robot capabilities.

We’ve been exploring the potential of using Gemini 4 Argon to accelerate research @Google. One of our tests was using videos of humans performing fun tasks (game of bags, anyone?) to autonomously spin up training simulations, rather than relying entirely on manual programming. https://t.co/3oK74wKSmt

We’ve been exploring the potential of using Gemini 4 Argon to accelerate research @Google. One of our tests was using videos of humans performing fun tasks (game of bags, anyone?) to autonomously spin up training simulations, rather than relying entirely on manual programming. https://t.co/3oK74wKSmt

Gemini 4 Argon is now available worldwide as Google's most capable frontier model

Google is rolling out Gemini 4 Argon worldwide to developers, businesses, and consumers, replacing Gemini Pro as its frontier series. Access includes the API through AI Studio and the Gemini platform, plus selected subscription tiers across Antigravity, the Gemini app, and Google Search's AI mode; free and Plus users are limited to Flash models. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with a 95 percent discount on cached inputs.

Gemini 4 Argon reaches GA, matching GPT-6 Astra on Artificial Analysis Intelligence Index

Artificial Analysis reports that Gemini 4 Argon is now GA, matching GPT-6 Astra on its Intelligence Index at 60% of the Cost per Task under discounted pricing until November 30. Argon costs $1.99 per task and $2/$10 per 1M input/output tokens, rising to $4/$20 after November 30, with a 95% cache-read discount. Artificial Analysis also reports substantial agentic evaluation improvements and a 15% hallucination rate on AA-Omniscience, the lowest among models scoring 45+ on its Intelligence Index.

Google releases Gemini 4 Argon with competitive pricing and claimed benchmark gains

Google launched Gemini 4 Argon, claiming improved reasoning and benchmark performance against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5. It is rolling out today to Pro and Ultra subscribers in the Gemini app and AI mode in Search, and Ultra subscribers on Antigravity; access for Plus subscribers and free users remains unclear. Introductory pricing is $2 per million input tokens and $10 per million output tokens, doubling at the end of November, though lower token rates do not necessarily mean lower task costs.

The same story is shown in the picks as 《Gemini 4 Argon is now available worldwide as Google's most capable frontier model》

Liquid AI releases LFM2.5-24B-A2B for fast, private on-device agents

Liquid AI released LFM2.5-24B-A2B, an open-weight reasoning model with 24B total parameters and 2B active parameters, for on-device agents that reason and call tools. Its context window expands from 32K to 131K tokens, and both the reasoning and Base versions are available on Hugging Face. Liquid AI reports 184 tokens/s on an Apple M5 Max and competitive tool-use results, while acknowledging that larger models retain advantages on knowledge-heavy and coding benchmarks.

Google releases Gemini 4 Argon with paid access and introductory API pricing

Baoyu reports that Google released Gemini 4 Argon on October 13 for complex, multistep tasks, citing Sundar Pichai’s announcement of global paid availability. Access includes the paid Gemini API and platform, Ultra subscriptions in Google Antigravity, and Pro or Ultra subscriptions in the Gemini app and Search. Introductory API pricing is $2 per million input tokens and $10 per million output tokens, including thinking tokens, through November 30; prices double December 1. Input, thinking, and output share a 1,048,576-token limit.

Google DeepMind launches Gemini 4 Argon for complex reasoning and agentic tasks

Google DeepMind has launched Gemini 4 Argon globally through its API and selected Google products, with access depending on subscription or service tier. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off and a 1M unified input-output token limit through the Gemini API. Google reports improvements across coding, professional workflows, scientific research, and robotics, alongside stronger agent safeguards and prompt injection resistance.

Google DeepMind Introduces Gemini 4 Argon for Robotics

Google DeepMind introduces Gemini 4 Argon, a general-purpose physical reasoning model for robot control, simulation reconstruction, physical understanding, and self-improvement. Google DeepMind reports an 87% ERQA score using pure reasoning and a simulation-generated training example that raises real-world robot success from 56% to 75% while reducing completion time from 35s to 24s. Argon is available today to Google AI Ultra subscribers through Google AI Studio and Google Antigravity, and through the paid Gemini API tier using gemini-4-argon in the Interactions API.

Artificial Analysis introduces Decision Index to benchmark decision models against language models

Artificial Analysis introduced Decision Index v1.0 to compare decision models’ capability, cost, and speed across classification, tool calling, knowledge, and reasoning. Jev 1.13 leads decision models with 58, while Claude Haiku 5.5 (Low) scores 61 but takes 1.4 seconds per task versus 0.13 seconds and costs $184 per 1M Tasks versus $42. Leading decision models perform comparably to language models on classification and tool calling but trail on knowledge and reasoning, where their answer probabilities generally exceed their accuracy.

Google rolls out Gemini 4 Argon across Gemini API, Gemini App, and Antigravity

Logan Kilpatrick announces that Gemini 4 Argon is rolling out globally today in the Gemini API, to Pro + Ultra users in Gemini App, and to Ultra subscribers in Antigravity. He says over 100,000 people at Google tested the model over the last month, with cyber defenders in the fairwind program and early access customers receiving it over the last 2 weeks. He highlights writing, knowledge, persistence for coding, and 1M token output when the model needs to work more.

The same story is shown in the picks as 《Gemini 4 Argon is now available worldwide as Google's most capable frontier model》

Google’s Gemini 4️⃣ Argon ✨ is now rolling out globally

Gemini 4️⃣ Argon ✨ is now rolling out globally. Google describes its frontier model as built for complex, real-world tasks across coding, finance, legal, cybersecurity, and science. See how to use Argon in your favorite Google product 🧵 https://t.co/PtoPqbtPel

The same story is shown in the picks as 《Gemini 4 Argon is now available worldwide as Google's most capable frontier model》

Google begins rolling out Gemini 4 Argon to paying users

Google is beginning a wider rollout of Gemini 4 Argon in the Gemini app and Antigravity today, limited to paying users, with incomplete access on lower-tier plans. Its capabilities are unchanged from the earlier announcement, and Google describes safeguards including malicious-instruction refusals and an automated misalignment monitor. Google also reports internal uses, including identifying redundant YouTube emoji rendering whose removal reduced server load by 45 percent and finding Project Wing simulator bugs dating to 2018.

The same story is shown in the picks as 《Gemini 4 Argon is now available worldwide as Google's most capable frontier model》

Build long-horizon agents with Grok 4.7 on Amazon Bedrock

AWS demonstrates how to build and evaluate a tool-calling agent using Grok 4.7 on Amazon Bedrock, comparing four reasoning effort levels across three synthetic tasks. The agent answered 67 of 72 runs correctly; xhigh achieved 18 of 18, while low achieved 15 of 18 using about an eighth of the output tokens. The guide covers task design, automated grading, production monitoring, and operational controls, noting that these small-scale results are indicative rather than statistically conclusive.

Shing-Tung Yau launches Chinese AI mathematics alliance amid Western skepticism

Shing-Tung Yau and leading Chinese mathematicians launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence on Sunday in Beijing, according to the South China Morning Post. It unites eight institutions and aims to connect over 1,000 researchers, with plans for specialized models, machine-verifiable proofs, and transparent AI reasoning. Meanwhile, 25 Fields Medal winners, including Terence Tao and Peter Scholze, worry that AI-generated proofs erode mathematical understanding and encourage dependence on tools researchers cannot follow.

Meta launches Muse Wonder and Muse Wonder Full frontier models

Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full, its second-generation model family for personal and professional agentic work. Latent Space reports that Wonder is available through the Meta Model API and Muse Code, while Full is limited to Muse Code subscribers. Artificial Analysis gives Wonder Full and Wonder Intelligence Index v4.3 scores of 56 and 52, respectively, versus Opus 5.5’s 58; Wonder supports a 1M context window and text, image, video, and audio inputs, with text-only output.

GPT-6 Astra helps mathematics enthusiast nor prove five circle-packing cases

QbitAI reports that mathematics enthusiast nor used GPT-6 Astra to prove optimality for packing 15, 16, 17, 18, and 20 unit disks inside a circle, completing the remaining cases for N≤20. The approach combines geometric constraints, finite computational enumeration, and exact optimality arguments. Nor describes substantial human involvement, including guiding the model, fixing issues, manually verifying validity, and rechecking the work with Astra and Opus 5.5.

The author suggests fitness hardware for Codex and Claude Code

I think there should be companion fitness hardware for Codex and Claude Code. For example, reasoning intensity could be tied to your grip strength: the stronger your grip, the higher the reasoning intensity. That way, you could work out harder to get higher reasoning intensity.

Sakana AI's self-improving model gets better at solving and judging, but can't catch its own blind spots

Researchers from Sakana AI and UC Berkeley introduced MASS, a method that uses copies of Qwen3.6-27B to solve tasks, evaluate results, and optimize workflows without a stronger teacher model. After two rounds, its test-task win rate against the base model reached 69.9 percent, although benchmark gains largely reflected more completed work and some scores declined. Shared blind spots allowed flawed results to pass evaluation and enter training, prompting the researchers to recommend independent checks, limits on autonomy, and clear intervention conditions.

Meta releases Muse Wonder and Muse Wonder Full with higher benchmark scores

Meta has released Muse Wonder and Muse Wonder Full, according to Artificial Analysis in the post quoted by Alexandr Wang. At max reasoning effort, they score 52 and 56 on the Artificial Analysis Intelligence Index v4.3, compared with Muse Spark 1.3’s 48, following a release gap of 6 weeks. Muse Wonder is available through Meta’s first-party API and Muse Code, while Muse Wonder Full is limited to Muse Code users; both have 1M-token context windows.

Meta releases frontier models Muse Wonder and Muse Wonder Full

Meta’s Superintelligence Lab released Muse Wonder and Muse Wonder Full on Monday, larger models described as more capable than Muse Spark. Meta reports benchmark scores comparable to Fable 5.1 and GPT-6 Astra and emphasizes information retention through “active memory.” Wonder costs $4.00 per million input tokens and $20.00 per million output tokens through the Meta Model API; Wonder Full is available only to Muse Code subscribers. Both models will power the Muse agent.

Meta launches Muse Wonder and Muse Wonder Full AI models

Meta launched Muse Wonder and Muse Wonder Full, available to Muse Code subscribers starting today; Wonder is also accessible through the Meta Model API. Meta claims internal benchmark testing shows the models can rival OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5 and Claude Fable 5.1. The company plans to power its Muse AI agent with Wonder, starting with subscribers, but has provided no timeline.

Microsoft announces Maia 300 AI chip to reduce reliance on Nvidia GPUs

Microsoft announced Maia 300, its third-generation in-house AI chip, with production rollout starting in 2027 and plans to make Maia its primary AI inference system. Microsoft expects 40 percent more tokens per megawatt than leading GPU-based systems and up to 7x Maia 200’s rack-scale token throughput in a full 96-accelerator configuration. Built on TSMC’s N2P process node, Maia 300 separates compute, memory controller, and I/O functions through a new chiplet architecture.

Microsoft introduces Maia 300 and plans multi-gigawatt Maia deployments

Microsoft is introducing Maia 300, which Satya Nadella says delivers 40% more tokens per watt compared to leading GPU-based systems. Starting next calendar year, Microsoft will scale its Maia family to multi-gigawatt deployments, making it the primary inference system for its own products and services. Microsoft is already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot, with a goal of building on full-stack efficiency gains while continuing to bring ecosystem innovations into its fleet.

Andrej Karpathy describes a widening gap in how people experience AI capabilities

Andrej Karpathy describes four tiers of LLM exposure, arguing that public perceptions lag behind capabilities experienced by frontier-system users. He estimates roughly six billion people have barely encountered LLMs, one to two billion use casual chatbot products, 20 million use frontier models professionally in math and code, and 5,000 access internal frontier systems. He claims those internal systems coordinate thousands of agents on projects that would take experts years, while human review falls behind, and attributes progress in coding and math to reinforcement learning on verifiable rewards and commercial training priorities.