Skip to content

#模型发布

Today 0 items
Aug 21Fri
  1. Hugging Face Blog66

    Up to 3.2x Faster Inference with LFM2.5-DSpark

    LiquidAI released DSpark draft model checkpoints for the LFM2.5 series' LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B, speeding up inference via speculative decoding without changing output quality; throughput on H100 improves up to 3.18x, and up to 2.87x on edge devices.

    Why it matters: The official release provides measured speedup data for the DSpark speculative decoding draft models along with llama.cpp and SGLang usage, so edge deployers can compare and judge the benefits.

  2. Microsoft Research50

    Broadening access to Skala creates a faster path to predictive DFT

    Microsoft Research released Skala 1.1, trained on 2.5 times the data of the previous version, ranking first in 32 of the 55 categories of the GMTKN55 benchmark with a weighted average error of 2.8 kcal/mol at compute cost comparable to a meta-GGA functional. Skala is now available in CP2K and being integrated into Psi4, FHI-aims, ORCA, and VASP, alongside a dynamic benchmark tracking each version's computational performance.

Aug 14Fri
  1. Google DeepMind72

    Introducing Gemini 3.7 Flash

    Google DeepMind released Gemini 3.7 Flash, positioned as its flagship model for coding and agents, just three weeks after Gemini 3.6 Flash.

    Why it matters: The official announcement includes multiple benchmark scores and the half-price pricing details, so readers can decide whether to switch to 3.7 Flash.

Aug 12Wed
  1. Google DeepMind74

    Putting sign language AI into users’ hands

    Google DeepMind released a large-scale multilingual sign language-to-text (SL2T) translation model, bringing new sign language input features to deaf and hard-of-hearing users, first supporting American Sign Language (ASL) to English in Gboard and Live Transcribe on Pixel 11, with more devices and languages to follow.

    Why it matters: The article gives concrete details on training scale, benchmark scores, and privacy design, letting readers understand the real technical path to deploying sign language translation.

  2. Google Research65

    Advancing AMIE towards expert-level audio-visual clinical consultations

    Google Research has launched AMIE (Video), built on Gemini and Project Astra, which can conduct real-time synchronized clinical video consultations, perceive non-verbal cues, and guide virtual physical exams.

    Why it matters: The article covers the architecture, evaluation scale, and quantitative comparisons against primary care physicians (PCPs), while candidly noting limitations like simulated scenarios, helping readers understand the practical boundaries of audio-visual clinical AI.

Aug 11Tue
Aug 10Mon
  1. Hugging Face Blog78

    Meta is back with Muse Glimmer: local, agentic, multimodal, and open source

    Meta today released Muse Glimmer, a 30B-parameter open-source multimodal model distilled from Muse under the Apache 2.0 license, aimed at privacy-focused local deployment and agentic uses like coding, document analysis, and personal assistance.

    Why it matters: The article provides architecture details, agent benchmark comparisons, and day-0 usage across inference stacks, helping readers evaluate local multimodal deployment options.

Aug 6Thu
  1. Google DeepMind79

    WeatherNext: AI model achieves breakthrough in forecasting cyclones

    Google DeepMind published a Nature paper and open-sourced WeatherNext 2, WeatherNext Cyclones, and WeatherNext 2-mini; a single AI model can simultaneously predict a tropical cyclone's track, intensity, and wind structure, offering on average about 24 hours more lead time than previous models — roughly a decade's worth of meteorological progress over the past 20 years.

    Why it matters: Google detailed the roughly one additional day of lead time in cyclone forecasting and released the code and weights, so researchers can directly reproduce or improve the models.

Aug 4Tue
  1. Mistral AI56

    Introducing Shieldstral.

    Mistral AI has released Shieldstral, a 3B open-weight multimodal safety classifier under the Apache 2.0 license. It frames moderation as binary question answering, with policies supplied as natural language questions at inference time without retraining, uniformly covering moderation, refusal detection and toxicity detection across text, images and prompt-response pairs; in official evaluations it performs on par with open guard models up to 7x its size and runs on a single 16GB GPU.

Jul 30Thu
  1. Google DeepMind67

    Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration

    Google DeepMind released Gemini Robotics ER 2, positioned as a high-level embodied reasoning model for robots that can orchestrate low-level control interfaces such as VLA models and call tools like Google Search.

    Why it matters: Official benchmark numbers on progress classification and temporal grounding, plus multi-robot collaboration demos, let readers evaluate how usable this embodied reasoning model is for real-world robot orchestration.

Jul 28Tue
  1. Google DeepMind65

    Gemini Robotics 2 brings whole body intelligence to robots

    Google DeepMind released Gemini Robotics 2, comprising three models — a VLA model, the Gemini Robotics ER 2 embodied reasoning model, and On-Device 2 that runs locally — enabling whole-body control, dexterous manipulation, and multi-robot collaboration for humanoid robots.

    Why it matters: The official release explains the division of roles among the three models, the data cost of adapting to new robots, and safety benchmarks, letting readers judge the path to real-world deployment of robotics AI.

Jul 21Tue
  1. Google DeepMind76

    Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

    Google DeepMind released three Gemini models: Gemini 3.6 Flash cuts output token usage by 17% versus 3.5 Flash on the Artificial Analysis Index, priced at $1.50/1M input and $7.50/1M output, with DeepSWE 49% vs 37% and OSWorld-Verified 83.0% vs 78.4%.

    Why it matters: The official article gives concrete pricing, token efficiency, and multiple benchmark numbers for the three new models, and explains the deployment scenarios each targets.

Jul 17Fri
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 8Wed
  1. Mistral AI49

    Introducing Robostral Navigate

    Mistral has released its first embodied navigation model, Robostral Navigate, an 8B-parameter model built fully in-house and trained on simulation data. Using only a single ordinary RGB camera, with no LiDAR or depth sensors, it can navigate according to natural language instructions.

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
Jun 30Tue
  1. Google Research61

    Introducing TabFM: A zero-shot foundation model for tabular data

    Google Research has launched TabFM, a foundation model for tabular data classification and regression that frames tabular prediction as an in-context learning (ICL) problem, producing predictions in a single forward pass without manual training, hyperparameter tuning, or feature engineering.

    Why it matters: The article explains TabFM's architecture, synthetic-data training approach, and TabArena benchmark results, letting readers weigh it against traditional tuning workflows.

Jun 23Tue
  1. Mistral AI70

    Introducing Mistral OCR 4

    Mistral released the OCR 4 document parsing model, which beyond text extraction returns bounding boxes, typed block classification and per-word confidence scores, supports 170 languages across 10 language groups, and can be fully self-hosted via a single container.

    Why it matters: Sourced from Mistral officially, it covers OCR 4's structured output, multilingual support and self-hosting options, and also discloses benchmark scoring limitations, serving as a selection reference.

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.

Jun 9Tue
  1. Google DeepMind76

    Introducing Gemma 4 12B: a unified, encoder-free multimodal model

    Google DeepMind released Gemma 4 12B, a unified, encoder-free multimodal model in which vision and audio inputs go directly into the LLM backbone, positioned between E4B and the 26B MoE.

    Why it matters: The article details the encoder-free architecture, 16GB-memory runnability, and Apache 2.0 open sourcing, helping readers evaluate on-device deployment options.

May 22Fri
  1. Mistral AI70

    Remote agents in Vibe. Powered by Mistral Medium 3.5.

    Mistral released the 128B open-weight flagship model Mistral Medium 3.5 (public preview), combining instruction following, reasoning and coding, with a 256k context window and a SWE-Bench Verified score of 77.6%. API pricing is $1.5 per million input tokens and $7.5 per million output tokens.

    Why it matters: The original provides model specs, benchmark scores, pricing and the open-weight access point, useful for evaluating it for self-hosting and asynchronous coding use cases.

May 18Mon
  1. Google DeepMind82

    Introducing Gemini Omni

    Google DeepMind launched Gemini Omni Flash, the first model in the Gemini Omni family, which can combine image, audio, video, and text inputs to generate videos, with multi-turn editing via natural language and support for character consistency and physical plausibility.

    Why it matters: The official introduction of the Omni family details the capability boundaries of conversational video editing, multi-input composition, and physical realism.

May 16Sat
  1. Google DeepMind84

    Gemini 3.5: frontier intelligence with action

    Google DeepMind has released the Gemini 3.5 model family, starting with 3.5 Flash, positioned for agents and coding. It outperforms Gemini 3.1 Pro on Terminal-Bench 2.1 (76.2%), GDPval-AA (1656 Elo), MCP Atlas (83.6%), and CharXiv Reasoning (84.2%), with output speed 4x that of other frontier models.

    Why it matters: The official release shares 3.5 Flash's benchmark scores, availability channels, and plans for a Pro follow-up, letting readers evaluate its trade-offs for agent and coding workloads.

Apr 16Thu
  1. Google DeepMind60

    Gemini 3.1 Flash TTS: the next generation of expressive AI speech

    Google DeepMind released the text-to-speech model Gemini 3.1 Flash TTS, which scores an Elo of 1,211 on the Artificial Analysis TTS leaderboard and supports 70+ languages and native multi-speaker dialogue.

    Why it matters: The official post introduces audio tags and AI Studio scene control among other new capabilities, showing readers how expressive control of TTS speech works in practice.

Apr 13Mon
Mar 24Tue
  1. Mistral AI65

    Speaking of Voxtral

    Mistral AI released its first text-to-speech model, Voxtral TTS, with 4B parameters supporting 9 languages including English, French and German, positioned for low-latency speech generation. In human evaluations its naturalness beats ElevenLabs Flash v2.5 and matches ElevenLabs v3; latency is 70ms for a 10-second sample plus 500 characters, with RTF around 9.7x.

    Why it matters: The original gives key details on latency, pricing, human evaluation comparisons and the open license, helping readers assess its fit for voice agent scenarios.

Mar 17Tue
  1. Mistral AI71

    Introducing Mistral Small 4

    Mistral AI released Mistral Small 4, the first small model to unify Magistral reasoning, Pixtral multimodal and Devstral agentic coding capabilities into a single model, open sourced under the Apache 2.0 license.

    Why it matters: The official release provides full architecture parameters, benchmark comparisons and deployment requirements, letting readers assess whether it can replace a multi-model stack.

Feb 5Thu
  1. Mistral AI65

    Voxtral transcribes at the speed of sound.

    Mistral AI has released Voxtral Transcribe 2, comprising two speech transcription models, Voxtral Mini Transcribe V2 and Voxtral Realtime, supporting 13 languages, speaker diarization, and word-level timestamps.

    Why it matters: The official post provides latency, error rate, and pricing data for both transcription models and releases the Realtime weights, so readers can compare against their current speech solutions.

Dec 9Tue
  1. Mistral AI72

    Introducing: Devstral 2 and Mistral Vibe CLI.

    Mistral AI has released the open source coding model family Devstral 2, comprising the 123B Devstral 2 (modified MIT license) and the 24B Devstral Small 2 (Apache 2.0). Both support 256K context windows and score 72.2% and 68.0% on SWE-bench Verified respectively.

    Why it matters: The official post covers both models' open source licenses, SWE-bench Verified scores, API pricing, and hardware requirements, so readers can assess local or on-device deployment feasibility.

Dec 3Wed
  1. Mistral AI78

    Introducing Mistral 3

    Mistral has released its new Mistral 3 model series, including the MoE-architecture Mistral Large 3 (41B active, 675B total parameters) and Ministral 3 models in 3B, 8B, and 14B sizes, all open sourced under Apache 2.0.

    Why it matters: The official announcement fully lists model specs, open source licenses, deployment options, and available platforms, so readers can assess feasibility of running them in their own environments.

Jul 30Wed
Jul 15Tue
  1. Mistral AI70

    Voxtral

    Mistral released Voxtral, an open-source speech understanding model available in a 24B Voxtral Small and a 3B Voxtral Mini, both under the Apache 2.0 license and callable via its API at prices as low as $0.001 per minute.

    Why it matters: Official benchmarks and API pricing for two open-source speech understanding model sizes let readers assess whether Voxtral can replace their current ASR solutions.

Jul 11Fri
Jun 10Tue
  1. Mistral AI68

    Magistral

    Mistral AI released its first reasoning model, Magistral, in two versions: the open-source Magistral Small (24B, Apache 2.0) and the enterprise Magistral Medium.

    Why it matters: With both open-source and commercial versions, official AIME2024 scores, open licensing and deployment channels, readers can use this to evaluate model selection.

May 28Wed
  1. Mistral AI56

    Codestral Embed

    Mistral AI has released Codestral Embed, its first code-specific embedding model, claiming it outperforms Voyage Code 3, Cohere Embed v4.0, and OpenAI's large embedding model on real-world code retrieval.

May 21Wed
  1. Mistral AI69

    Devstral

    Mistral AI and All Hands AI have launched Devstral, an agentic large language model for software engineering tasks, released as open source under the Apache 2.0 license.

    Why it matters: The official release includes model weights, API pricing, specific SWE-Bench Verified scores and deployment requirements, letting readers assess fit for local or enterprise coding scenarios.

May 7Wed
  1. Mistral AI64

    Medium is the new large.

    Mistral AI released Mistral Medium 3, positioned as a new class of model with SOTA performance at 8x lower cost. It reaches or exceeds 90% of Claude Sonnet 3.7 on benchmarks, is priced at $0.4 input / $2 output per M token, surpasses Llama 4 Maverick and Cohere Command A, and can be self-deployed on 4 or more GPUs.

    Why it matters: The official release provides benchmarks against Claude Sonnet 3.7, pricing, and deployment requirements, letting readers evaluate its cost-effectiveness for enterprise adoption.

Mar 17Mon
  1. Mistral AI62

    Mistral Small 3.1

    Mistral AI released Mistral Small 3.1, which improves text performance over Mistral Small 3, adds multimodal understanding, extends the context window to 128k tokens, and runs at 150 tokens per second.

    Why it matters: Coming from the official release, it includes benchmark comparisons, the open-source license, and hardware requirements, so readers can assess its fit for lightweight deployments.