Tripo AI has launched EdenLive, a real-time generative renderer dedicated to interactive scenes, as part of its interactive environment world model Project Eden, and it is now open-sourced to developers. EdenLive decouples scene evolution logic from visual rendering; rendering can be started with just a 2D/3D scene or reference video, a target first frame, and text instructions, and it supports both 480p and 720p resolutions; the 5-billion-parameter version can achieve real-time generation at over 24 frames per second at 720p on a single consumer-grade RTX 5090. The open-source release includes the 5-billion and 14-billion parameter model weights, real-time video generation code, the EdenGolem v0.1 skill for converting text into simple games, and a complete prompt-to-game front-end toolchain; the code is released on GitHub under the Apache 2.0 license, and the weights are available on Hugging Face.
Tripo AI has launched EdenLive, a real-time generative renderer dedicated to interactive scenes, as part of its interactive environment world model Project Eden, and it is now open-sourced to developers. EdenLive decouples scene evolution logic from visual rendering; rendering can be started with just a 2D/3D scene or reference video, a target first frame, and text instructions, and it supports both 480p and 720p resolutions; the 5-billion-parameter version can achieve real-time generation at ov
Anthropic ran an experiment with 50 causal inference experts; the agent's preference at each decision point was randomly assigned by a coin flip, yet participants still adopted the agent's preference in 86% of analytical decisions, with flip rates of 62% to 82%. Afterward, 52% believed the final recommendation was more like their own judgment, 35% believed it was basically or entirely their own, and only 15% mentioned AI. The study also found that after using the agent, the rate of serious analytical errors dropped from 55% (6/11) to 6% (3/50), and the median time required fell from 90 minutes to 65 minutes, but the no-AI group had a smaller sample and slightly lower expertise.
Anthropic ran an experiment with 50 causal inference experts; the agent's preference at each decision point was randomly assigned by a coin flip, yet participants still adopted the agent's preference in 86% of analytical decisions, with flip rates of 62% to 82%. Afterward, 52% believed the final recommendation was more like their own judgment, 35% believed it was basically or entirely their own, and only 15% mentioned AI. The study also found that after using the agent, the rate of serious analy
OpenAI released the GPT-6 Luna (October 29) update, which responds more concisely and reduces overly sycophantic phrasing; under the Preparedness Framework it is still rated High for both cybersecurity and biochemical capabilities, and it uses the same set of safeguards as GPT-6 Luna (October 7). This version improves over the previous version on the gore evaluation in Production Benchmarks, achieving a Pareto improvement between safely handling harmful requests and helpfulness; in the U18 evaluation it improves on sexual content, eating disorders, emotional dependence, and gore, and in the instruction hierarchy evaluation it reaches 99.91% robustness.
OpenAI released the GPT-6 Luna (October 29) update, which responds more concisely and reduces overly sycophantic phrasing; under the Preparedness Framework it is still rated High for both cybersecurity and biochemical capabilities, and it uses the same set of safeguards as GPT-6 Luna (October 7). This version improves over the previous version on the gore evaluation in Production Benchmarks, achieving a Pareto improvement between safely handling harmful requests and helpfulness; in the U18 evalu
Google has announced that it will gradually discontinue Jules, the AI coding assistant launched in December 2024, and apply the experience gained from the tool to Antigravity, its agent-first development platform. Jules's official X account said that Jules was originally an experiment to understand how asynchronous coding agents can help developers ship faster, and that in the future it will merge remote background task execution with an agent-centric development environment into a unified experience. Antigravity was launched by Google DeepMind on November 18, 2025, and is a standalone IDE distributed as a VS Code fork, supporting Windows, macOS, and Linux.
Google has announced that it will gradually discontinue Jules, the AI coding assistant launched in December 2024, and apply the experience gained from the tool to Antigravity, its agent-first development platform. Jules's official X account said that Jules was originally an experiment to understand how asynchronous coding agents can help developers ship faster, and that in the future it will merge remote background task execution with an agent-centric development environment into a unified exper
Ant inclusionAI has released the ZwZ model family (4/7/8B) accompanying its ICML 2026 paper, achieving SOTA on multimodal perception benchmarks among open-source models, while requiring only a single forward inference pass and no tool calls at inference time. The method is Region-to-Image Distillation (R2I), which distills high-quality VQA supervision generated from enlarged local regions back into the full image, with 37K training samples. It also introduces ZoomBench, containing 845 VQA items covering 6 fine-grained perception dimensions, on which Qwen2.5-VL-7B achieves only 42.5% average accuracy.
Ant inclusionAI has released the ZwZ model family (4/7/8B) accompanying its ICML 2026 paper, achieving SOTA on multimodal perception benchmarks among open-source models, while requiring only a single forward inference pass and no tool calls at inference time. The method is Region-to-Image Distillation (R2I), which distills high-quality VQA supervision generated from enlarged local regions back into the full image, with 37K training samples. It also introduces ZoomBench, containing 845 VQA items
Ant Group's GuangJian team has open-sourced DefakerOne2 on GitHub, a 2B-parameter unified visual forensics model built on the Qwen3.5-2B multimodal backbone, covering AI-generated content, DeepFake faces, document tampering, natural image tampering, face and ID anti-counterfeiting, and AI-generated video detection. The model was trained on approximately 50 million samples, employs a three-stage Tide-Style Tuning process from LoRA to full parameters and back to LoRA, and uses entropy-based test-time routing to select between the video and frame pathways. The team claims that it achieves SOTA on 40 of 53 detection benchmarks and on 11 of 14 localization benchmarks, and that the weights and the RRVBench benchmark have been released on ModelScope.
Ant Group's GuangJian team has open-sourced DefakerOne2 on GitHub, a 2B-parameter unified visual forensics model built on the Qwen3.5-2B multimodal backbone, covering AI-generated content, DeepFake faces, document tampering, natural image tampering, face and ID anti-counterfeiting, and AI-generated video detection. The model was trained on approximately 50 million samples, employs a three-stage Tide-Style Tuning process from LoRA to full parameters and back to LoRA, and uses entropy-based test-t
OWASP ranks the failure to decommission non-human identities as the top risk in its 2025 Top 10, and AWS IAM user access keys do not expire automatically—forgotten keys remain valid until someone manually disables or deletes them. The article presents a three-step governance plan: use CloudQuery queries to find keys that have been unused for more than 90 days and are still active, use OPA policies to block the creation of long-term credentials during the Terraform/OpenTofu plan phase, and set a finite lifecycle for each key.
OWASP ranks the failure to decommission non-human identities as the top risk in its 2025 Top 10, and AWS IAM user access keys do not expire automatically—forgotten keys remain valid until someone manually disables or deletes them. The article presents a three-step governance plan: use CloudQuery queries to find keys that have been unused for more than 90 days and are still active, use OPA policies to block the creation of long-term credentials during the Terraform/OpenTofu plan phase, and set a
Amazon releases five Alexa devices: Echo Dot priced at $99.99, with a built-in smart home hub and eero Wi-Fi extender; Echo Studio with Home Camera priced at $249.99; Echo Time smart bedside clock priced at $199.99; Alexa Photo Frame priced at $199.99; Echo Show 16 pre-order priced at $499.99, with the Nanomatte screen version at $569.99, shipping later this year.
Michael Pfaffenberger of Pydantic used 5 small Pydantic AI applications and 8 models to conduct 5,276 scored runs, comparing code mode with direct tool calling. On long tasks, code mode reduced average cost by 58%, latency by 51%, and increased success rate by 16 percentage points; however, for short tasks with 5 items, 3 out of 5 tasks were instead 21% to 38% more expensive. The author argues that models are better at writing loops than at acting as the loop, and the gap widens as tasks get longer.
Michael Pfaffenberger of Pydantic used 5 small Pydantic AI applications and 8 models to conduct 5,276 scored runs, comparing code mode with direct tool calling. On long tasks, code mode reduced average cost by 58%, latency by 51%, and increased success rate by 16 percentage points; however, for short tasks with 5 items, 3 out of 5 tasks were instead 21% to 38% more expensive. The author argues that models are better at writing loops than at acting as the loop, and the gap widens as tasks get lon
NVIDIA has released the open-weight model Nemotron 3.5 Transcribe on Hugging Face and ModelScope, integrating transcription, speaker diarization, translation, and audio question answering into a single checkpoint. It supports transcription in 40 languages, translation between 31 languages and English, and speaker labeling in 13 languages, can process recordings up to 1 hour long, and can label up to 8 speakers. The model ranks first among open-weight models on the English leaderboard of the Hugging Face Open ASR Leaderboard with an average WER of 5.33%, and achieves 2.8% AA-WER on Artificial Analysis with approximately 4B activated parameters. The simultaneously released Granary v2 covers 48 languages and 463,062 hours of speech data, and the accompanying NeMo Speech fine-tuning recipe reduced WER on held-out speakers from 22.44% to 12.17% in a medical case.
NVIDIA has released the open-weight model Nemotron 3.5 Transcribe on Hugging Face and ModelScope, integrating transcription, speaker diarization, translation, and audio question answering into a single checkpoint. It supports transcription in 40 languages, translation between 31 languages and English, and speaker labeling in 13 languages, can process recordings up to 1 hour long, and can label up to 8 speakers. The model ranks first among open-weight models on the English leaderboard of the Hugg
In issue 254 of "New Gadgets," the Shaoshupai editorial team shared four products they recently purchased or received for testing. The HomePod mini 2's chip has been upgraded from the S5 to the S10, and Wi-Fi has been upgraded to Wi-Fi 6, but the difference in everyday listening is extremely small compared with the previous generation; under the 27.2 system, AirPlay latency is greatly improved. The Sonos Beam Ultra achieves 7.1.2 Dolby Atmos on its own, and the Sonos 27 system provides a native cloud MCP service that lets commonly used AI directly control speaker playback.
In issue 254 of "New Gadgets," the Shaoshupai editorial team shared four products they recently purchased or received for testing. The HomePod mini 2's chip has been upgraded from the S5 to the S10, and Wi-Fi has been upgraded to Wi-Fi 6, but the difference in everyday listening is extremely small compared with the previous generation; under the 27.2 system, AirPlay latency is greatly improved. The Sonos Beam Ultra achieves 7.1.2 Dolby Atmos on its own, and the Sonos 27 system provides a native
Meesho said on its Q2 FY2026-27 earnings call that AI agents have become its default way of developing software: one engineer leading a group of agents can take on the work of what used to be a three-person team, and 65% of teams are driven by a single engineer. The company said its internal productivity score rose 171% from March to September 2026, code review approval time fell 41%, and the time for a model to enter online testing shrank from 17 days to 3 days. At the same time, AI has also pushed up costs: GPU capacity increased 150%, and server costs as a share of net merchandise value (NMV) rose from 1.9% to 2.3%. The company plans to migrate some workflows to cheaper or open-weight models. Meesho also made clear it will not add a platform fee at the shopping cart stage, even though it loses about 2.4 rupees per order.
Meesho said on its Q2 FY2026-27 earnings call that AI agents have become its default way of developing software: one engineer leading a group of agents can take on the work of what used to be a three-person team, and 65% of teams are driven by a single engineer. The company said its internal productivity score rose 171% from March to September 2026, code review approval time fell 41%, and the time for a model to enter online testing shrank from 17 days to 3 days. At the same time, AI has also pu
Factory releases Droid 1.0, the largest update yet to its agent framework, enabling programmatic tool calling (PTC) by default, which lets the model write JavaScript scripts to execute hundreds of tool calls within a single request in a sandbox. On Terminal-Bench 4, when paired with GPT-6.1 Sol, Droid scores 3.5 points higher than Codex at 12% lower cost; when paired with Claude Opus 5.5, it scores 3.0 to 7.6 points higher than Cursor, OpenCode, Devin, and Claude Code, at up to 24% lower cost. Droid 1.0 is available today with macOS and Windows desktop versions and a terminal UI, and also offers the Droid SDK (@factory/droid-sdk).
Factory releases Droid 1.0, the largest update yet to its agent framework, enabling programmatic tool calling (PTC) by default, which lets the model write JavaScript scripts to execute hundreds of tool calls within a single request in a sandbox. On Terminal-Bench 4, when paired with GPT-6.1 Sol, Droid scores 3.5 points higher than Codex at 12% lower cost; when paired with Claude Opus 5.5, it scores 3.0 to 7.6 points higher than Cursor, OpenCode, Devin, and Claude Code, at up to 24% lower cost. D
Replit has launched Agent Habitat, allowing agents to access enterprise knowledge and systems within boundaries the company can control, building security governance around four principles: least privilege, fault isolation, auditable production changes, and traceable delegated work. Its credential broker retrieves credentials from the vault and injects them only after each request is authorized, so credentials never enter the agent's context; enterprise connector management supports SAML SSO, SCIM, and role-based permissions. Replit said the kernel vulnerability disclosed last week did not affect its existing isolation environment, and it has concentrated on patching tools and deploying firewall protection.
Replit has launched Agent Habitat, allowing agents to access enterprise knowledge and systems within boundaries the company can control, building security governance around four principles: least privilege, fault isolation, auditable production changes, and traceable delegated work. Its credential broker retrieves credentials from the vault and injects them only after each request is authorized, so credentials never enter the agent's context; enterprise connector management supports SAML SSO, SC
Replit releases ViBench 2.0 to evaluate AI agents' ability to build a complete product feature by feature starting from an empty project, comprising 17 products, 241 feature requests, and 544 test plans. GPT-6.1 Sol leads with a 65% pass@1, GPT-6 Astra is at 60%, and Opus 5.5 is at 49%, with all three passing over 90% of test plans. Opus 5.5 completes a build in about 70 minutes on average, roughly 2.5 times faster than Sol; DeepSeek V4.1 Flash passes 77% of test plans at about $2.50 per build, but no open-weight model can pass a complete build.
Replit releases ViBench 2.0 to evaluate AI agents' ability to build a complete product feature by feature starting from an empty project, comprising 17 products, 241 feature requests, and 544 test plans. GPT-6.1 Sol leads with a 65% pass@1, GPT-6 Astra is at 60%, and Opus 5.5 is at 49%, with all three passing over 90% of test plans. Opus 5.5 completes a build in about 70 minutes on average, roughly 2.5 times faster than Sol; DeepSeek V4.1 Flash passes 77% of test plans at about $2.50 per build,
a16z's survey of 100 CISOs at companies with more than 1,000 employees shows that 87% expect cybersecurity spending to increase, driven mainly by AI-related threats and the expanded attack surface from enterprises' own AI deployments. All seven of the most-watched security initiatives for 2027 are AI-related; SOC automation received the most first-priority votes, but only 18% have put AI SOC tools into production. 45% plan to make new purchases or upgrades within 12 months for internal AI agent security, and 42% plan to procure offensive security capabilities.
a16z's survey of 100 CISOs at companies with more than 1,000 employees shows that 87% expect cybersecurity spending to increase, driven mainly by AI-related threats and the expanded attack surface from enterprises' own AI deployments. All seven of the most-watched security initiatives for 2027 are AI-related; SOC automation received the most first-priority votes, but only 18% have put AI SOC tools into production. 45% plan to make new purchases or upgrades within 12 months for internal AI agent
Google DeepMind has released Gemini 4 Carbon, which ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59 and leads in 4 of 9 evaluations including HLE-Diamond and TB-Science. The model focuses on cybersecurity and long-horizon tasks, claiming 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and is initially available only to trusted cybersecurity personnel through the Fairwind Program. In terms of specifications, it supports a 1M token context and 1M maximum output, with input covering text, images, video, and speech, and output in text; Artificial Analysis lists a standard price of $4/$20 per million input/output tokens, plus a 50% launch discount.
Google DeepMind has released Gemini 4 Carbon, which ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59 and leads in 4 of 9 evaluations including HLE-Diamond and TB-Science. The model focuses on cybersecurity and long-horizon tasks, claiming 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and is initially available only to trusted cybersecurity personnel through the Fairwind Program. In terms of specifications, it supports a 1M token context and 1
AnySpeechRecognizer supports engine switching on top of the Apple SpeechAnalyzer API, with options including NVIDIA Parakeet TDT 0.6B v3 (supporting 25 European languages), Moondream post-trained Parakeet Ultra, the English version Parakeet Unified, and Alibaba Qwen3-ASR (0.6B/1.7B, covering 30 languages and 22 Chinese dialects). These models have been converted to Core ML and published on the Hugging Face Hub; Parakeet takes about 20–40 seconds to transcribe 1 hour of audio, while Qwen3-ASR is slower and requires several GB of memory. The package also adds a speaker diarization module, SpeakerDiarizer, which the Apple framework lacks; it is based on NVIDIA Streaming Sortformer and can track up to 4 speakers.
AnySpeechRecognizer supports engine switching on top of the Apple SpeechAnalyzer API, with options including NVIDIA Parakeet TDT 0.6B v3 (supporting 25 European languages), Moondream post-trained Parakeet Ultra, the English version Parakeet Unified, and Alibaba Qwen3-ASR (0.6B/1.7B, covering 30 languages and 22 Chinese dialects). These models have been converted to Core ML and published on the Hugging Face Hub; Parakeet takes about 20–40 seconds to transcribe 1 hour of audio, while Qwen3-ASR is
The Baseten engineering team, through native audio decoding, KV cache FP8 quantization, and reduced host-device synchronization, increased batch transcription throughput for Qwen3-ASR 1.7B on a single RTX PRO 6000 by 7x and reduced latency by 6.9x. After optimization, the transcription cost for one hour of audio is approximately $0.0013, i.e., 1/10 of a cent per audio hour, with no significant change in word error rate. The optimized version has been made the default deployment in the Baseten Model Library and can be called through an OpenAI-compatible transcription client.
The Baseten engineering team, through native audio decoding, KV cache FP8 quantization, and reduced host-device synchronization, increased batch transcription throughput for Qwen3-ASR 1.7B on a single RTX PRO 6000 by 7x and reduced latency by 6.9x. After optimization, the transcription cost for one hour of audio is approximately $0.0013, i.e., 1/10 of a cent per audio hour, with no significant change in word error rate. The optimized version has been made the default deployment in the Baseten Mo
Ben's Bites author Ben Tossell published a post reflecting on his struggle of being overwhelmed by AI tools and multiple agents and repeatedly overturning his direction, and announced that for the next 30 days he will only use agents for things he finds interesting, will not introduce new tools that don't solve real pain points, will not summarize long articles, and will not do things for the sake of making money. He also mentioned that Google released Gemini 4 Carbon and the world model Genie 4, as well as a test in which experts followed an agent's random choices 86% of the time.
Ben's Bites author Ben Tossell published a post reflecting on his struggle of being overwhelmed by AI tools and multiple agents and repeatedly overturning his direction, and announced that for the next 30 days he will only use agents for things he finds interesting, will not introduce new tools that don't solve real pain points, will not summarize long articles, and will not do things for the sake of making money. He also mentioned that Google released Gemini 4 Carbon and the world model Genie 4
