All AI updates

Anthropic's Claude Fable 5.2 doesn't generate images, it paints them

Anthropic's Claude Fable 5.2 creates images by writing Python programs of 1,000 to 2,000 lines that drive a simulated paint engine, placing individual brushstrokes on a virtual canvas rather than using diffusion. Developer Jake Eaton showed a photorealistic parking garage scene made with roughly 700,000 simulated brushstrokes, and engineer Michael Graczyk shared a 21-second video of a bird at sunrise built from exactly 111 strokes, with no undo and deliberate wobble in the lines. Anthropic reports 64.9 percent for Fable 5.2 on its internal RoboDojo Sim benchmark versus 29.8 percent for OpenAI's GPT-6 Astra, though researcher Aris Zhu's real robot-arm comparison found Astra painted faster and was the only model to stand a domino on end.

Bland Launches Tofu-1, a Full-Duplex Speech-to-Speech Model for Calls

Bland, the San Francisco voice-AI company, has released Tofu-1, a full-duplex speech-to-speech model in research preview that targets calls from moving cars, hospital rooms and weak-signal lines, listening while it speaks and reasoning while it talks. Bland lists speaker tracking, noise separation and paralinguistics as its audio skills, and says callers can interrupt or correct it mid-sentence while it keeps the conversation moving and reports results only when work is done. Access is a closed enterprise research preview for banking, healthcare, insurance and government, with Bland citing January availability and no published size, latency, pricing or benchmarks.

Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU

Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter mixture-of-experts model for autonomous agent and coding tasks that activates roughly 48 billion parameters per token and was trained entirely on Chinese hardware. It handles up to one million tokens of text, image, and video input via API, and Meituan reports 88.4 on Terminal Bench v2.1 and 64.6 on DeepSWE, close to but below Claude Opus 5 and GPT-5.6 Sol on several benchmarks. The weights are MIT-licensed, and the self-hosting guide describes a 194-chip Huawei Ascend Atlas A2 cluster with no Nvidia GPU guidance.

Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol

Google's Gemini 4 Carbon became the first non-OpenAI model in a while to top the MathArena math benchmark, displacing GPT-6.1 Sol, with Claude Fable 5.2 taking second place. Its most common failure mode is exceeding the time limit, and in one case it ran a script for more than eleven hours without checking its progress. With the reasoning settings used in testing, Carbon and Fable 5.2 cost roughly 15 and 40 times as much as GPT-6.1 Sol, respectively.

Runway's open-source Praxis-1 turns video AI into a robot controller

Runway has released Praxis-1, a World Action Model that takes camera images, text instructions and robot position data to predict scene changes and derive robot movements, with weights and code under the Apache 2.0 license. Built on a video model fine-tuned with human action videos and robot data, it scored 46.2 percent on Nvidia's RoboLab-120 leaderboard, ahead of FLUX 3 Action at 42.9 percent, though Runway ran the tests itself and they have not been independently verified. The model still needs specialized robot data for adaptation, sometimes confuses similar-looking objects, and cannot reliably plan complex action sequences on its own.

Perplexity Releases Decider v1.3: Open-Weight Decision Models Add Locate

Perplexity released pplx-decider-v1.3 and pplx-decider-flash-v1.3, open-weight decision models that return probabilities for yes/no, multiple-choice and score questions rather than generated text, with weights under Apache 2.0 on Hugging Face and availability in the Perplexity Decisions API and on OpenRouter. The main addition is Locate, which extracts exact text spans by type or instruction for NER, PII detection and extractive QA. Perplexity reports 91.6% mean accuracy on 63 query-classification tasks and a 0.27% answer-flip rate, with all figures self-reported.

Nil-3B Open Source Sovereign AI

This repository hosts the nil-3b-q4_k_m.gguf file for running the NIL-3B project locally with a GGUF-compatible inference runtime, via a chat script that expects the file in the project root. It also documents an uploader script that uses a Hugging Face token and repo ID from a .env file to create the model repository if needed and upload the GGUF file. The root LICENSE covers only the project source code and does not by itself confirm license or redistribution rights for the model weights or training data.

Google Releases Gemini 4 Hydrogen: 2.5x Faster Than Gemini 3.8 Flash

Google released Gemini 4 Hydrogen, a closed-weights model from Google DeepMind for coding and agentic work, with a 1,048,576-token unified window, four thinking levels (default high, plus a new xhigh), and API pricing of $1.50/$7.50 per 1M input/output tokens. Google reports it scores 51 on the Artificial Analysis Intelligence Index at high, 57% on Terminal-Bench 4.1 versus 18.2% for Gemini 3.8 Flash, and finishes tasks 2.5x faster than 3.8 Flash. It is available via the Gemini API, AI Studio, Android Studio and Google products, with Antigravity access requiring Google AI Pro or Ultra.

Featured coverage of this story: 《Introducing Gemini 4 Hydrogen》

llm-gemini 0.37

Google launched Gemini 4 Hydrogen, the lowest-cost member of the Gemini 4 family, priced at $1.50/M input, $0.075/M cached input and $7.50/M output, compared with Carbon and Argon at $2.00/$0.10/$10.00 through Nov 30, 2026, doubling from Dec 1, 2026. The post notes Carbon and Argon share the same price, that Hydrogen is only a 25% discount until December 1st (then 75%), and that it remains far more expensive than Claude Haiku 5.5 and GPT-6 Luna at $0.10/$0.50. Hydrogen is described as the new free tier model and the model powering AI features in Google Search.

Edge0-145B: Frontier AI Capabilities on Your Own Computer

The company open-sourced Edge0-145B and Edge0-120B, describing Edge0-145B as the first open-source 145-billion-parameter on-device model, alongside a unified inference framework and deployment tools. It reports 29GB peak memory, 80+ tokens/s decoding, 1,100+ tokens/s prefill, roughly 120W typical power and 128K–256K context on a compatible laptop, with benchmark scores compared against named cloud models. The release also describes shared-matrix sparse experts, tiered GPU/system-memory/SSD loading and CPU-GPU co-scheduling, plus local demonstrations in video, Blender 3D scenes, games, drone control and PCB design.

Liquid AI Releases LiquidCodec, a 74M-Parameter Streaming Audio Codec

Liquid AI released LiquidCodec, a ~74M-parameter streaming speech tokenizer used in its LFM3-Audio-2B and LFM3-Omni-2B models, which converts 24 kHz audio into 12 token streams at 12.5 Hz frames and back. Liquid AI reports best ViSQOL and lowest mel distance on all 8 test sets versus X-Codec 2, NanoCodec and Mimi, with encode RTF 0.0028 and decode RTF 0.0115 on a single CPU thread, though NanoCodec scored higher PESQ (WB) on 7 of 8 sets. Weights are on Hugging Face, with Q4_0, Q4_K_M and Q8_0 GGUF builds for llama.cpp.

Google’s new Gemini model is better at understanding complicated questions

Google says its new Gemini 4 Hydrogen model can give more helpful answers to complex questions, including ones with multiple parts, hard-to-understand typos, or highly specific requests, by breaking them down and addressing each part. It is available starting today in Search, the Gemini app, all tiers of Google AI Studio, the Gemini API, Gemini for business, and Google Antigravity for AI Pro and Ultra subscribers, and is now the default model in AI Mode. Google says it delivers significant leaps in speed and intelligence compared to Gemini Flash 3.8 and matches or outperforms competing models like Claude Sonnet 5.5 and GPT-6.1 Terra.

Featured coverage of this story: 《Google Search gets a major upgrade with Gemini 4 Hydrogen》

Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch

Google released Gemini 4 Hydrogen, a fast general-purpose model for developers and everyday users that scores 51 on the Artificial Analysis Intelligence Index at its default "high" thinking level (54 at the new "xhigh" tier), beating the larger Gemini 4 Argon's 50. It has a one-million-token input and output window and costs $1.50 per million input tokens and $7.50 per million output tokens, with Google reporting a jump from 18.2% to 57% on Terminal-Bench 4.1 versus Gemini 3.8 Flash. Artificial Analysis found Hydrogen hallucinates only 8% of the time but lags on agentic terminal work, scoring 54% on Terminal-Bench 4.0 and 50% on Terminal-Bench-Science 0.1.

Featured coverage of this story: 《Introducing Gemini 4 Hydrogen》

Google’s latest Search upgrade promises better answers to your most complicated questions

Google says its new Gemini 4 Hydrogen model, described as its "biggest model improvement to Search yet," improves Search's handling of multi-part questions, highly specific requests and confusing typos. Hydrogen completes tasks 2.5 times faster than Gemini 3.8 Flash and is rolling out to all Search users globally, while also coming to AI Studio, the Gemini API, the Gemini App, AI Pro and Ultra subscribers, and Antigravity.

Featured coverage of this story: 《Introducing Gemini 4 Hydrogen》

Google rolls out Gemini 4 Hydrogen, finally replacing 3.8 Flash

Google has released Gemini 4 Hydrogen, which replaces Gemini 3.8 Flash across its products with immediate access and more relaxed limits than earlier Argon and Carbon rollouts. Google says Hydrogen runs at four thinking levels (low, medium, high, and new xhigh), is 2.5 times faster than Flash 3.8 even in high mode, cuts "fabricated claims" by more than 5x, and more than triples 3.8 Flash's Terminal-Bench score. It scored 54 on the Artificial Analysis Intelligence Index in xhigh mode versus 50 for Argon.

Featured coverage of this story: 《Introducing Gemini 4 Hydrogen》

LLMON: An LLM-Native Markup Language, now in Granite 4.1 3B and 8B

IBM researchers introduce LLMON (LLM Object Notation), an XML-like markup language that uses reserved special-token delimiters to explicitly mark instructions, data, requirements and execution bindings in an LLM's context, addressing the brittleness of JSON-style formats for causal models. They release two experimental preview models derived from Granite 4.1 base checkpoints (granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview) plus a GitHub toolkit, reporting that on the OPI benchmark LLMON models avoid 47% more indirect prompt injection attacks than the best frontier model tested while maintaining general capability on par with Granite 4.1 Instruct. The models ship with their own chat template and lossless JSON-to-LLMON converters, and the post notes the previews are for exploration, not production.

Introducing Gemini 4 Hydrogen

Google announced Gemini 4 Hydrogen, a fast, efficient model for everyday coding and agentic workflows, available across the Gemini API, AI Studio, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro and Ultra subscribers in Google Antigravity. It is priced at $1.50 per million input tokens and $7.50 per million output tokens across its 1 million-token context window, and surpasses Argon on the Artificial Analysis Intelligence Index at its default high thinking level (51 vs. 50, reaching 54 at xhigh). Google also introduced Orchestrator in Google Antigravity, which switches between models like Carbon and Hydrogen within a session and reached near-Carbon task success rates on Terminal-Bench 4.0 while consuming 24% less quota and completing 19% faster.

ElevenLabs Scribe v3 Live uses conversational context to stop voice agents from cutting off callers mid-sentence

ElevenLabs released Scribe v3 for pre-recorded audio and Scribe v3 Live for low-latency real-time use, both available through the ElevenAPI and ElevenAgents platform. The company says Scribe v3 cuts word errors by about 25 percent versus its predecessor across 102 languages, while Artificial Analysis benchmarks confirm top-tier accuracy but note slower speed and mid-pack pricing. Scribe v3 Live analyzes conversational context alongside the audio signal to detect when a speaker has actually finished, and both models support preloaded specialized terms, natural-language formatting rules, and mid-sentence language switching.

Google Deepmind shifts genomics AI from bulk tissue predictions to single cell type resolution with AlphaGenome 2

Google DeepMind released AlphaGenome 2, a conditional genomics model trained on more than 107 million individual cell profiles from 68 human and mouse tissues that predicts variant effects for specific cell types rather than across many tissues at once. On the TenK10K benchmark of 28 primary immune cell types it identified about 30 percent of causal variants and correctly predicted up- or down-regulation in 43 percent of cases, versus 16 and 23 percent for a fine-tuned AlphaGenome 1. It is available for noncommercial research via API, with commercial access through a closed preview, and is not approved for clinical diagnostics.

DeepL unveils Aure, its most lifelike Voice AI translation model yet

DeepL announced Aure, a new Voice AI model powering DeepL Voice that continuously adapts translated speech to reflect shifts in tone, emotion, emphasis and pace, building on the company's earlier voice preservation technology. Aure analyzes incoming speech continuously using both translated words and source audio, and DeepL says it combined training techniques separating voice identity, prosody and spoken content with reinforcement learning. The capabilities apply automatically across DeepL Voice for Online Meetings, In-Person Conversations and the DeepL API for Voice, initially supporting English, German, Spanish, French, Italian, Japanese, Portuguese and Korean.

Gemini Decisions API

Google has joined Jev and OpenAI in the decision model category with "Gemini Decision 1", an API that supports the same three question types as Jev (yes/no, choice, and rubric-scoring) and reuses Jev's term "Noul" for yes/no questions. It is priced like GPT-6 Luna at 10 cents per million input tokens with free output, while Jev remains cheapest at 4.2 cents per million; unlike Jev (text and JSON only) and OpenAI (text and images), Gemini Decision 1 covers text, images, audio, video, and PDFs. Testing it on a video returned probability values rather than the 1.0 or 0.0 yes/no answers Jev and OpenAI give.

Google DeepMind Releases AlphaGenome 2 for Cell-Type-Specific Variant Effects

Google DeepMind has released AlphaGenome 2, a conditional sequence-to-function DNA model that reads a 1 Mb DNA window plus a target cell state to predict variant effects in specific cell types, adapting to unseen single-cell datasets without fine-tuning. On TenK10K single-cell eQTLs it recovered 30% of causal variants at 90% precision versus 16% for fine-tuned AlphaGenome 1, with direction-of-effect recall rising from 23% to 43%. It is available through the AlphaGenome API for non-commercial research, with weights to come upon final publication.

Google joins the decision-model wave with Gemini Decision 1, scoring 1,000 questions in a single pass

Google DeepMind is introducing the Gemini Decisions API, built for fast closed-ended decisions such as yes/no questions, multiple choice, and scale-based ratings, powered by the Gemini Decision 1 model that processes up to 1,000 independent questions in a single pass across text, images, audio, video, and PDFs. It scores options directly and returns probabilities with confidence values in 150 to 250 milliseconds, and Google says it already uses the API internally for dynamic model routing in Gemini Notebook, design prototyping in Stitch, and safety checks in Google Search. The API costs $0.10 per million input tokens with no output-token charge, about five times cheaper than Gemini 3.5 Flash-Lite according to Google, and is available now through Google AI Studio with integrations for OpenRouter, LangChain, Vercel, and other platforms rolling out.

Featured coverage of this story: 《Gemini Decisions API: Fast, calibrated multimodal decisions》

AlphaGenome 2: Deciphering how cells read DNA to better understand disease

Google DeepMind announced AlphaGenome 2, a conditional DNA sequence model trained on over 107 million individual cell profiles across 68 tissues that predicts how genetic variants affect gene expression in specific human cell types and states. The model identified nearly twice as many causal variants (30% vs. 16%) and directional gene effects (43% vs. 23%) as the original AlphaGenome fine-tuned on single-cell data on the TenK10K immune cell dataset, and can generalize to unseen single-cell experiments without retraining. It is available for non-commercial research via the AlphaGenome API and in private preview for commercial research.

Google’s AlphaGenome 2 zooms in on the cellular roots of disease

Google DeepMind unveiled AlphaGenome 2, a successor to last year's AlphaGenome that predicts how genetic variants affect individual cell types and states, trained on more than 100 million cell profiles across 68 tissues. Google says the model identifies nearly twice as many variants that can change gene expression as its previous state-of-the-art model, and can make predictions about new cell types and states without retraining. It is available for non-commercial research via the AlphaGenome API from today, with a limited private preview for commercial research.

DeepL wants real-time translation to finally capture sarcasm and mood shifts

DeepL's new Aure model for DeepL Voice transfers tone, emotion, emphasis, and speaking pace from the original speaker into the target language by continuously analyzing both the translated text and original audio, unlike the earlier approach that used a static voice sample. Reinforcement learning keeps accents out of the output, and Aure runs automatically in DeepL Voice for Online Meetings, In-Person Conversations, and the DeepL API, launching with support for English, German, Spanish, French, Italian, Japanese, Portuguese, and Korean. The article also notes DeepL's earlier voice preservation rollout in September, its May restructuring that cut about 250 jobs, its acquisition of the Mixhalo team, and Block's Dessa speech model for detecting emotions and sarcasm.

MBZUAI’s IFM Releases PAN 2, an Open 29B World Model for Physical AI

MBZUAI's Institute of Foundation Models released PAN 2, an Apache 2.0 world model that takes a scene and text action and simulates the outcome as video, shipping weights, training code and inference code. It pairs an LLM planner with a 29B MMDiT diffusion transformer trained from scratch, outputting 704×1248 video at 24 fps up to 20 seconds. IFM reports 34.0% full-plan completion on 100 multi-step tasks, leading six tested models, while trailing MiniMax-H3 on single actions and human preference.

Fish Audio Releases Live 3, a Full-Duplex Voice Model That Reacts in 186 ms

Fish Audio has released Live 3, a full-duplex voice model for real-time voice agents that keeps listening while it speaks and learns turn-taking from real human conversations rather than hand-tuned pipeline rules. Fish Audio says it can decide to respond in as little as 186 ms for short reactions like "mm-hmm," though no benchmark scores or latency distribution have been published. It is available on the Fish Agent Platform, via the Fish Audio API, and on-premises for enterprises, priced at $0.035 per minute until December 31, 2026 (then $0.07), and supports English only.

BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model

Tether AI Research introduces BotPsy, a 0.8B-parameter model built on Qwen3.5-0.8B for function calling and agentic workflows, suitable for on-device deployment. It roughly doubles its base model's tool-use scores (BFCL v4 23.4 to 44.0, ACEBench 30.1 to 58.7, τ-bench 23.2 to 51.5) and outperforms every model up to 1.2B parameters tested on all six agentic benchmarks, while also beating Qwen3.5-2B-Think on five of the six. It will be released with open weights under Apache 2.0 in bf16, Q8 and Q4 as safetensors and GGUF files, and is intended for low- to medium-difficulty tasks, not high-stakes use.

Ant Group's Robbyant open-sources tactile AI model for robot hands

Ant Group's Robbyant has open-sourced a tactile AI model for robot hands, according to TechNode. The article is available at the linked URL.

llm-gemini 0.36

Gemini 4 Carbon, announced two weeks ago for "a subset of cyber defenders", is now available to everyone at launch prices of $2/million input and $10/million output, scheduled to double to $4/$20 on December 1st. The post shows three pelican test images for low, medium and high thinking efforts (defaulting to medium) and an Antigravity IDE test producing a Three.js animation of Yonkers as a mech stomping around Manhattan, Queens and Brooklyn using OpenStreetMap data. An update notes a YouTube demo in which Carbon's /swarm feature built an open-world 3D game exploring Google's Mountain View campus as a pelican riding a GBike.

NaiveAI Promises 2,100 Tokens per Second. How Well Does It Code?

Latent Space benchmarks NaiveAI's open-weights Naive-N0.5-Flash, a MiMo-V2.5 derivative adapted to DeepSeek Sparse Attention with about 3.25T extra training tokens, against DeepSeek V4.1 Flash 1017 and GLM 5.3 Flash on speed, verbosity and three app-building prompts. It reports peak throughput near 2,100 tokens/s and a median response time of 3.1 seconds versus 5.3 and 10.1 seconds, but far higher thinking-token use on hard questions and mean times similar to DeepSeek. In the apps, Naive-N0.5-Flash produced the best solar system, tied DeepSeek on the tree model and trailed it on ray-tracer efficiency, while its no-thinking variant failed to build a usable tree app.

Gemini 4 Carbon arrives just in time to feel like yesterday's news in the fastest model cycle yet

Google has released Gemini 4 Carbon, which it says complements rather than replaces Gemini 4 Argon, targeting complex multi-step coding, debugging and open-ended scientific research. Anthropic's newly unveiled Fable 5.2 outperforms Carbon in the Artificial Analysis test even at lower reasoning intensity, though Carbon costs far less per task. Google also previewed a "/swarm" feature in Antigravity and reported that Carbon achieved a zero percent prompt injection success rate on the Gray Swan IPI benchmark.

Introducing Scribe v3 and v3 Live, our most accurate speech to text models

ElevenLabs launched Scribe v3 for recorded audio, which makes 27% fewer word errors than Scribe v2 across 102 languages and has the lowest diarization error of any model tested, and Scribe v3 Live for realtime audio, which makes 43% fewer semantic word errors than Scribe v2 Realtime and adds speaker diarization and improved turn detection. Scribe v3 Live is available today in ElevenAgents, and both models are available in ElevenAPI, where switching from Scribe v2 is a one-line change to the model ID (scribe_v3 or scribe_v3_live). Both run in the US by default with isolated EU, India, and Singapore environments, Zero Retention Mode, and SOC 2, ISO 27001, PCI DSS Level 1, and HIPAA-eligible coverage.

Google DeepMind's Gemini Robotics 2.5 scores 96 percent sorting metal parts in a Mercedes-Benz assembly pilot

Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model that converts visual input and language instructions into robot movements, reporting 30 to 100 percent higher success rates and 20 to 30 percent faster task completion than its own baselines across more than 200 tasks. In a pilot with Apptronik and Mercedes-Benz on active production lines in Kecskemét, Hungary, and Sindelfingen, Germany, the Apollo 2 humanoid sorted metal parts from bulk bins into trays, achieving a reported 96 percent success rate. DeepMind also published the open-source ASIMOV-Jailbreak benchmark on Hugging Face and said safe task completion improved by 33 percent near obstacles and 43 percent near people.

Google releases Gemini 4 Carbon with agent ‘swarm’ feature

Google is rolling out Gemini 4 Carbon to paying developers and enterprises via the Gemini API in Google AI Studio and Gemini for business, and to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Antigravity, after an earlier release to select cybersecurity defenders. Google says Carbon is built for tasks that test a model's limits, such as complex multi-step coding and open-ended scientific research, and is previewing a 'swarm' feature on Antigravity that turns a single prompt into a coordinated team of agents. It launches at an introductory $2 per million input tokens and $10 per million output tokens, with prices doubling at the end of November.

Featured coverage of this story: 《Gemini 4 Carbon is now available》

Gemini 4 Carbon is now available

Google released Gemini 4 Carbon to developers, enterprises, scientists, and consumers, describing it as its strongest model for cybersecurity capabilities and long-horizon agentic workflows in knowledge work, software engineering, and scientific reasoning. It ranks #1 on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench at more than 2X lower cost per task than the next best external model, and is available via the API in Google AI Studio and Gemini for business at $2 per million input tokens and $10 per million output tokens. Google also previewed /swarm in Google Antigravity and reported safety results including a 0% prompt injection attack success rate on the Gray Swan IPI benchmark.

Zyphra Releases ZONOS3-Mini: Open-Source TTS With 100M Active Parameters

Zyphra has released ZONOS3-Mini, an Apache 2.0 text-to-speech model that is a sparse Mixture of Experts with 800M total and about 100M active parameters, supporting zero-shot voice cloning and word-level delivery control. It runs via WebGPU in the browser, MLX on Apple Silicon, ggml on CPU/CUDA/Metal/Vulkan, and SGLang-Omni on NVIDIA GPUs, with Zyphra reporting 80 ms median time to first audio on one H100 at 8 concurrent requests. Zyphra reports it ranks first among open models on its self-run ZTTS2-Eval Capability Index at 66.2 in speaker-embedding mode, while reference-clip mode scores only 44.7.

Mitra-v2: Scaling synthetic pretraining for tabular foundation models

Mitra-v2 is a second-generation tabular foundation model for classification and regression, pretrained entirely on synthetic tables with 10x as many rows and roughly 3x as many columns as Mitra-v1, plus a new hybrid SCM prior and hybrid optimizer. On TabArena it leads the single-model field, beating Mitra-v1 on 32 of 33 datasets where both run natively and completing all 51 datasets within the one-hour budget, and fine-tuned Mitra-v2 ranks first on ScoringBench. The fine-tuned models and inference/fine-tuning code are released under Apache-2.0 on Hugging Face.

DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"

Google DeepMind unveiled Gemini Robotics 2.5, a vision-language-action model it says is much better at high-dexterity tasks, evaluated across more than 200 activities including serving food with tongs and using a pipette. The platform pairs Gemini Embodied Reasoning, which plans tasks from instructions and live video, with the VLA that generates robot actions; some Gemini robots did actual work in Mercedes-Benz factories. Google also showed Apptronik's Apollo 2 humanoid with Gemini Robotics 2.5 tying a bow with its five-fingered hands, and the team is preparing an open source safety benchmark for robots working alongside humans.