Skip to content

#推理

Today 0 items
Oct 16Fri
  1. Hugging Face Blog77

    Welcome Beam: a 501B open model you can own, customize, and scale

    Reflection has open-sourced its first model, Beam (reflectionai/Beam-1.0-501B-A23B), on Hugging Face: a 501B-parameter MoE model (23B activated per token) with 1M token context, released under the Apache 2.0 license and targeting coding and agent use cases.

    Why it matters: The original provides architecture details, benchmark results, hardware requirements, and fine-tuning recipes, so readers can assess the feasibility of deploying and customizing Beam on their own hardware.

Oct 15Thu
  1. Mistral AI60

    Introducing Timestral

    Mistral has released Timestral, its first family of time series foundation models in four sizes — 5M, 28M, 123M and 209M parameters — with the largest model ranking first among pretrained zero-shot models on the GIFT-Eval, TIME and fev-bench public benchmarks, beating the previous SOTA with about two-thirds the parameters.

    Why it matters: The article gives verifiable details on the four sizes, benchmark rankings and CPU inference latency, letting readers evaluate it for time series forecasting deployments.

  2. Hugging Face Blog64

    Apertus 1.5 Release Notes

    Project Apertus has released Apertus 1.5, adding image and audio understanding and an optional thinking mode to its 8B and 70B models via continued pretraining, and extending the context window to 262,144 tokens — 4x that of Apertus 1.0.

    Why it matters: The official release notes fully document the training pipeline, open data and intermediate checkpoints, letting readers trace how the multimodal and reasoning capabilities were built.

  4. Google Research74

    Analyzing technical documents with Deep Review

    Google Research has publicly released Deep Review, an agent system that provides rigorous feedback and technical verification for complex technical documents. Formerly the Paper Assistant Tool (PAT), it partnered with four conferences — STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 — to deliver pre-submission feedback on over 16,000 opted-in manuscripts.

    Why it matters: The article details real-world usage of over 16,000 manuscripts across four top conferences and the four-stage agent pipeline design, which can transfer to your own document review workflows.

Oct 14Wed
  1. Google DeepMind78

    Developing Gemini 4 Argon responsibly and safely

    Google DeepMind released Gemini 4 Argon, focused on complex tasks like coding, finance, law, cybersecurity and science, and opened it first to cyber defenders through the Fairwind Program.

    Why it matters: Google also details specific defenses against misuse, misalignment and prompt injection along with quantified results, letting readers see how frontier model safety is actually done.

  2. Hugging Face Blog67

    Run fast, private agents on-device with LFM2.5-24B-A2B

    Liquid AI released LFM2.5-24B-A2B, a reasoning model with 24B total parameters and only 2B activated per token, aimed at fast private on-device agents, published with open weights on Hugging Face.

    Why it matters: The release includes full benchmark comparisons, inference speed and a five-stage training pipeline, so readers can judge whether it fits local agent workflows.

  3. Google DeepMind83

    Gemini 4 Argon: Built to solve your hardest problems

    Google DeepMind released Gemini 4 Argon, available worldwide to developers, enterprises, scientists and consumers via the Gemini API, AI Studio, Antigravity and the Gemini app, priced at $2 per million input tokens and $10 per million output tokens, with cached input tokens at a 95% discount.

    Why it matters: The official release includes API pricing, engineering and science use cases and safety data, letting readers assess its real value for coding and agent scenarios.

Oct 13Tue
Oct 7Wed
  1. Hugging Face Blog73

    One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO

    The NVIDIA team, building on Nemotron 3, used SFT, RL, and a generate-verify-improve reasoning system to score 30/42 at IMO 2026 (gold threshold 29), while Nemotron-3-Ultra-CC scored 535.4/600 in unofficial IOI 2026 testing (gold threshold 361.12).

    Why it matters: The article lays out the complete fine-tuning and inference recipe from base model to gold medal, with open weights, data, and pipelines, so readers can reproduce this domain-specialization approach.

  2. OpenAI News66

    Sharing AI progress in mathematics

    OpenAI released new results on its internal frontier model solving open mathematical problems, and published Lean proof formalizations and research details on GitHub.

    Why it matters: The article reports that OpenAI shared results from its internal frontier model on open math problems and released Lean formalized proofs and research details on GitHub.

  3. Microsoft Research48

    What AI gets wrong and what failure teaches us

    In a Microsoft Research podcast, partner research manager Jennifer Neville speaks with applied scientist Chad Atalla about how evaluation drives AI systems past performance limits to meet user needs, and about the "unexpected failures" models show when tested beyond standard benchmarks. She shares practical advice on using existing AI systems and stresses the importance of carefully checking data when results don't match expectations.

Sep 30Wed
  2. AWS Machine Learning Blog69

    Bring near-Astra intelligence to everyday work with GPT-6.1 Sol on Amazon Bedrock

    GPT-6.1 Sol is now generally available on Amazon Bedrock, focusing on stronger reasoning for agent coding, computer use, and professional workloads. According to OpenAI, its DeepSWE v1.1 score matches GPT-6 Astra at about one-fifth the per-task cost and beats GPT-6 Sol's best score by 6.4 percentage points; it also outperforms GPT-6 Sol on transparency, user intent, and explicit constraint evaluations.

    Why it matters: The official announcement gives specific comparisons against GPT-6 Sol and GPT-6 Astra, helping readers evaluate its cost-effectiveness for agentic work scenarios.

Sep 29Tue
  1. Hugging Face Blog63

    NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction

    NVIDIA has released the open-source tabular foundation model Kumo Tabular, now on Hugging Face. Given labeled tables, the model predicts labels for new rows in a single forward pass with no training, tuning, or feature engineering; it supports classification and regression, is Transformer-based with column, row, and in-context attention, and was pretrained entirely on synthetically generated tables.

    Why it matters: The official post covers the architecture, pretraining recipe, and four benchmark results of a zero-finetuning tabular foundation model, letting readers judge whether it can replace traditional gradient-boosted-tree workflows.

  2. AWS Machine Learning Blog64

    Grok 4.7 is now available on Amazon Bedrock

    xAI's Grok 4.7 is now available on Amazon Bedrock, offering a 500K token context window and four reasoning effort levels: low, medium, high, and xhigh.

    Why it matters: The article covers how to access it on Bedrock, reasoning effort configuration, and cost trade-offs, so readers can directly plan their usage and spend.

Sep 26Sat
Sep 16Wed
Sep 9Wed
  1. OpenAI News69

    GPT-6 Astra: The next generation in intelligence for work

    OpenAI has released GPT-6 Astra, positioned as its most powerful business-focused model, with advanced reasoning and computer use capabilities, plus stronger writing and design judgment.

    Why it matters: The official launch of a business-oriented model highlights reasoning, computer operation, and writing/design judgment, letting readers assess where it fits in their workflows.

Sep 8Tue
  1. OpenAI News81

    On the Navier–Stokes Millennium Prize Problem

    OpenAI shared an AI-generated solution approach to the Navier–Stokes Millennium Prize Problem, including a solution write-up and a Lean formalized proof. The material comes from an official OpenAI announcement; the feed does not include the full text, so specifics of the proof and methods should be checked via the original link.

    Why it matters: OpenAI has publicly released an AI-generated approach to the Navier–Stokes Millennium Prize Problem with a Lean formalized proof, letting readers track AI's actual output in mathematical proof work.

Sep 1Tue
Aug 25Tue
  1. Hugging Face Blog61

    Granite 4.2 LLMs: How They're Built

    IBM released the Granite 4.2 reasoning model family, comprising 3B, 8B and 30B dense decoder models built on the Granite 4.1 base (pretrained from scratch on about 15T tokens, with a five-stage strategy extending context to 512K), post-trained with SFT and multi-stage GRPO reinforcement learning, all open-sourced under the Apache 2.0 license.

    Why it matters: The official disclosure covers the full training pipeline from pretraining and SFT to multi-stage GRPO and agentic RL, transferable to similar model-building workflows.

Aug 12Wed
  2. Microsoft Research58

    Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement

    Microsoft Research introduces CARE-X, a research model and unified chest X-ray VLM built on SigLIP2-so400M and Phi-4-mini-instruct (3.8B), jointly trained with auxiliary heads to provide report generation and structured predictions with confidence, and aligned to clinical rewards using DAPO reinforcement learning.

Aug 4Tue
Jul 23Thu
  1. Google Research49

    Towards a quantum computer that learns from its errors

    Google Research published a reinforcement learning-based quantum error correction control framework in Nature, letting an agent continuously tune thousands of control parameters using error detection events to counter drift during computation. In experiments injecting artificial drift on the Willow superconducting processor, RL control improved logical stability by 3.5x and further reduced the logical error rate by 20% on top of expert calibration.

Jul 15Wed
  1. Hugging Face Blog80

    Welcome Inkling by Thinking Machines

    Thinking Machines has released the open-source multimodal model Inkling on Hugging Face, with about 1T total parameters (975B, a MoE with 41B active), natively accepting image, text, and audio inputs, supporting 1M context, trained on 45 trillion tokens, and shipping BF16 and NVFP4 weights.

    Why it matters: The official blog provides architecture details, VRAM requirements for each tier, and deployment configurations, letting readers evaluate multimodal deployment costs and model selection.

Jul 10Fri
  1. Hugging Face Blog61

    Profiling in PyTorch (Part 3): Attention is all you profile

    Hugging Face published part three of its Profiling in PyTorch series, using the profiler on an NVIDIA A100 to compare traces of six attention implementations: naive attention, in-place masking, and SDPA math/efficient/flash/cuDNN.

    Why it matters: The original benchmarks kernel-level differences across six attention implementations with the profiler, teaching readers to read from traces how in-place saves a copy and what each SDPA backend really costs.

Jul 2Thu
  1. Mistral AI65

    Leanstral 1.5: Proof Abundance for All

    Mistral released the Apache-2.0 open-source model Leanstral 1.5, with 119B total parameters and 6B activated parameters, trained via mid-training, supervised fine-tuning and CISPO reinforcement learning.

    Why it matters: The official post discloses the full training pipeline, benchmark figures and real code verification cases, letting readers gauge how practical formal verification models are for mathematical proofs and open-source bug finding.

Jul 1Wed
  1. Berkeley AI Research33

    2026 BAIR Graduate Showcase

    The Berkeley AI Research (BAIR) lab releases its 2026 PhD graduate showcase, highlighting graduates such as Baifeng Shi, Charlie Snell, and Hanlin Zhu and their work in robotics and embodied intelligence, LLM reasoning and test-time scaling, generative models, AI safety and human-AI interaction, and speech and medical AI.

Jun 25Thu
Jun 11Thu
  1. Google DeepMind70

    DiffusionGemma: 4x faster text generation

    Google DeepMind released DiffusionGemma, an open-source experimental model that generates whole blocks of text via a text diffusion approach, achieving up to 4x inference speedups on dedicated GPUs (over 1,000 tokens/second on an NVIDIA H100 and over 700 tokens/second on an RTX 5090).

    Why it matters: The official release specifies the hardware behind the 4x speedup and output quality trade-offs, letting readers judge whether it suits local low-concurrency workloads.

May 29Fri
May 27Wed
May 8Fri
Apr 30Thu