64 releases covered live on the show — every model, product, paper and tool that mattered, with links and our analysis.
Pipecat releases Smart-Turn, an open source semantic VAD model
The Pipecat team (from Daily) released Smart-Turn, an open source semantic voice activity detection model that understands when a speaker has actually finished their turn rather than just detecting silence. Kwindla Kramer joined the show to break down how semantic VAD makes voice agent conversations feel far more natural, with a community training effort at turn-training.pipecat.ai.
Google ships Quantization-Aware Trained Gemma 3 models for consumer GPUs
Google released Quantization-Aware Training (QAT) versions of the Gemma 3 family, dramatically cutting memory requirements while preserving quality. The 27B model drops from a hefty 54GB to just 14.1GB, and even the 1B model goes from 2GB to about half a gig, making state-of-the-art open models runnable on consumer GPUs. Wolfram took the 4B QAT model for a spin in LM Studio on the show.
FramePack generates 120-second videos on just 6GB of VRAM
FramePack, from ControlNet creator Lvmin Zhang (lllyasviel), is an open source next-frame prediction approach for long video generation that runs on consumer hardware. It can generate videos up to 120 seconds long on as little as 6GB of VRAM by packing input frame context into a fixed length.
Nari Labs' Dia: a wild 1.6B open source TTS model that blew up Twitter
Nari Labs released Dia, a 1.6B parameter open-weights text-to-speech model that absolutely blew up Twitter with its expressive, emotional dialogue generation, including laughs, coughs, and multi-speaker conversations. Built by a tiny team, it punches far above its weight against commercial TTS systems and supports voice cloning, with demos available on Fal.ai.
NVIDIA releases DAM-3B for region-based image and video captioning
NVIDIA dropped the Describe Anything Model (DAM-3B), a 3 billion parameter multimodal model for region-based image and video captioning. You can point it at a specific region of an image or video and it generates a detailed description of just that area. NVIDIA also published an accompanying DescribeAnything dataset and a Hugging Face demo.
Sand AI surprises with MAGI-1, a 24B streaming autoregressive video model
Sand AI released MAGI-1, a 24B autoregressive diffusion model for long-form, streaming video generation with remarkable character consistency, often the Achilles' heel of AI video. It predicts video in 24-frame chunks with causal attention between them, enabling real-time streaming generation where compute doesn't scale with length. Nisten speculated it could be a major step toward usable AI-generated movies by solving the face/character consistency problem.
Tencent's Hunyuan 3D 2.5 jumps to 10B params with PBR textures and rigging
Tencent updated its 3D generation model to Hunyuan 3D 2.5, now boasting 10 billion parameters, up from 1B. They highlight massive leaps in precision with 1024-resolution geometry, high-quality textures with PBR support, and improved skeletal rigging for animation.
ByteDance publishes Seaweed-7B video generation foundation model
ByteDance publicly presented Seaweed-7B, a 7B parameter video generation foundation model, showing competitive video quality from a comparatively small model. Details and demos were published at seaweed.video.
ByteDance Seedream 3.0: bilingual 2K text-to-image model
ByteDance's Seed team announced Seedream 3.0, a powerful bilingual (Chinese/English) text-to-image model that generates native 2048x2048 images with fast inference of around 3 seconds for a 1K image on an A100. It challenges the top closed image generation models.
Cohere Embed 4: multimodal embeddings for enterprise search
Cohere released Embed 4, a multimodal embedding model aimed at enterprise search and retrieval over mixed text and image documents. It is available through Cohere's API.
DolphinGemma: Google's audio model for decoding dolphin communication
Google, with Georgia Tech and the Wild Dolphin Project, announced DolphinGemma, a ~400M parameter audio model based on the Gemma architecture using SoundStream audio tokenization. Trained on decades of recorded dolphin clicks, whistles and pulses, it aims to decipher structure in dolphin communication and runs on a Pixel phone for field deployment.
Google launches Gemini 2.5 Flash with controllable thinking budgets
Google answered OpenAI's launch week with Gemini 2.5 Flash, a fast reasoning model that introduces controllable thinking budgets so developers can dial how much the model reasons per request. It is available through the Gemini API and developer platform.
Kling 2.0 Creative Suite launches
Kuaishou's Kling AI launched Kling 2.0 along with a broader Creative Suite, upgrading its video generation model and tooling. The release kept up the rapid pace in the closed-source video generation race during a packed vision and video week.
Microsoft releases BitNet 1.58-bit model weights on Hugging Face
Microsoft published BitNet (listed in the show notes as BitNet v1.5), its native 1.58-bit quantized LLM, as open weights on Hugging Face. The ternary-weight approach targets extremely efficient CPU inference at a fraction of the memory of standard models.
OpenAI launches GPT-4.1 family (4.1, mini, nano) in the API
OpenAI released the GPT-4.1 family of models, available via API only, in three sizes: 4.1, 4.1-mini and 4.1-nano. The family features a 1M token context window, in contrast to o3's 200k, and is aimed at developers building on long-context and coding workloads.
OpenAI launches o3 and o4-mini, SOTA reasoning models with tool use
OpenAI shipped o3 and o4-mini in ChatGPT and the API, with o3 setting new SOTA records on Codeforces, SWE-bench, MMMU and more. For the first time the models can use tools (web search, Python, image generation) during the reasoning process, and they can think visually by cropping, zooming and rotating images. o3 scored $65k on the Freelancer eval versus o1's $28k, and o4-mini hits 99.5% on AIME with a Python interpreter.
Prime Intellect launches INTELLECT-2, a 32B globally-distributed RL run
Prime Intellect released INTELLECT-2, a 32B reasoning model trained with globally decentralized reinforcement learning, a follow-up to the INTELLECT-1 decentralized pretraining run covered on the show in December. The release includes open weights on Hugging Face, a tech report, and the PRIME-RL training code.
Z.ai (formerly chatGLM) releases the GLM-4-0414 open-source family
Z.ai, the rebranded Zhipu AI / chatGLM team, released the GLM-4-0414 family of open-source models. The drop includes base, reasoning and rumination variants published on Hugging Face and GitHub.
Amazon unveils Nova Sonic, a speech-to-speech foundation model
Amazon announced Nova Sonic, a foundational speech-to-speech model that unifies speech understanding and generation for real-time, natural-sounding voice conversations. It is available through Amazon Bedrock as part of the Nova family.
Deep Cogito debuts Cogito v1 Preview models from 3B to 70B, beating DeepSeek 70B
New lab Deep Cogito released the Cogito v1 Preview family of open models ranging from 3B to 70B parameters, claiming SOTA results at each size and beating DeepSeek's 70B distill. The models are available on Hugging Face, giving local AI enthusiasts the small-to-mid sizes Llama 4 skipped.
HiDream-I1-Dev: 17B MIT-licensed image model surpasses Flux 1.1 [pro]
HiDream released HiDream-I1-Dev, a 17B parameter open-weights image generation model under an MIT license. It became the new leading open-weights image generator, surpassing Flux 1.1 [pro] on quality benchmarks.
Jina Reranker M0: SOTA multilingual, multimodal document reranker
Jina AI released Jina Reranker M0, a state-of-the-art multimodal and multilingual document reranker model. It reranks documents that include both text and images, targeting retrieval and RAG pipelines, with weights available on Hugging Face.
Meta drops Llama 4 Scout (109B) and Maverick (400B) open-weights MoE models
Meta released the long-awaited Llama 4 family in a chaotic Saturday drop: Scout (17B active / ~109B total, 16 experts) and Maverick (17B active / ~400B total, 128 experts), with a 2T-parameter Behemoth still in training. The models are multimodal, multilingual MoE architectures trained on ~30T tokens with FP8 and interleaved attention (iRoPE), claiming 10M context for Scout and 1M for Maverick. The release was marred by drama: the LMArena version differed from the released model, and the community criticized the lack of small local-friendly sizes.
Moonshot drops Kimi-VL and Kimi-VL-Thinking, tiny A3B open vision models
Moonshot AI released Kimi-VL and Kimi-VL-Thinking, compact vision-language models with only ~3B active parameters (A3B MoE). The thinking variant adds reasoning to a tiny VLM, and both are available openly on Hugging Face.
NVIDIA ships Nemotron Ultra, a 253B pruned and distilled Llama 3.1-405B
NVIDIA released Nemotron Ultra, a pruned and distilled finetune of Llama 3.1-405B at roughly half the parameters (253B). Its benchmarks even included Llama 4 comparisons, showing the older finetuned Llama beating the new models on AIME, GPQA and more. It supports 128K context and fits on a single 8xH100 node for inference.
DeepCoder-14B: open RL-finetuned coder beats DeepSeek R1 and o3-mini on coding
Together AI and Agentica (UC Berkeley Sky Computing Lab) released DeepCoder-14B-Preview, a reasoning model finetuned with RL that beats DeepSeek R1 and even o3-mini on several coding benchmarks. The project aims to democratize RL: the team open-sourced the model, the training dataset, the Weights & Biases logs, and the eval logs. Guest Michael Luo from Agentica joined the show to discuss the release.
OpenHands LM 32B: MIT-licensed coding agent model hits 37.2% SWE-Bench
All Hands AI (formerly OpenDevin) released OpenHands LM 32B, an MIT-licensed Qwen finetune that scores 37.2% on SWE-Bench Verified, competing with much larger models on real-world repo tasks. The OpenHands agent also took the #2 spot on the new Live SWE-Bench leaderboard, and the 32B model runs locally on a single RTX 3090. A hosted OpenHands Cloud version is also available; guest Xingyao Wang joined the show to discuss it.
Gladia launches Solaria speech-to-text model
Gladia launched Solaria, a new speech-to-text model offered through its transcription platform. It arrived in a busy week for voice AI alongside Hailuo's Speech-02 TTS.
Dream 7B: a diffusion language model challenger unveiled
Researchers unveiled Dream 7B, a diffusion-based language model that posts strong benchmark results, notably on planning-style tasks like Sudoku, possibly because parallel generation handles global constraints better than autoregression. It hints at viable alternative LLM architectures, but the weights were not yet released at show time, so results could not be independently verified.
Nomic Embed Multimodal: SOTA embeddings for visual documents
Nomic AI released Nomic Embed Multimodal, new 3B and 7B parameter embedding models built on Alibaba's Qwen2.5-VL. They achieve SOTA on visual document retrieval by embedding interleaved text-image sequences, ideal for PDFs and complex webpages. The 7B model ships under Apache 2.0 with open weights, code, and data; guest Zach Nussbaum discussed the release on the show.
Runway Gen-4 announced with major gains in video consistency
Runway announced Gen-4, its next-generation video model focused on character and world consistency across shots. Example videos showed notably coherent characters and scenes, pushing AI video further toward usable filmmaking.
Character.AI opens early access to AvatarFX talking avatars
Character.AI announced AvatarFX, now in early access, which turns static images into speaking, emoting video avatars. It targets bringing characters to life for conversational and creative use cases.
Mistral releases Classifiers Factory
Mistral announced Classifiers Factory, a service for building and training custom text classifiers on its platform. Covered as a quick item in the Big CO LLMs + APIs section of the show.
Anthropic launches Max plan at $200/mo with higher usage quotas
Anthropic introduced a new Max subscription tier priced at $200 per month, offering significantly more usage quota than the standard Pro plan. It mirrors OpenAI's Pro-tier pricing strategy for power users.
Google launches Firebase Studio AI app-building environment at Cloud Next
As part of a flood of announcements at Google Cloud Next 2025, Google launched Firebase Studio, a browser-based AI-powered environment for building and shipping full-stack apps. It was one of the headline developer-facing launches from the event.
Amazon announces Nova Act browser agent SDK
Amazon entered the agent race with Nova Act, an agent designed to take actions in web browsers, possibly built with talent from the Adept acquisition. Amazon claims it beats Claude 3.5 and OpenAI's computer-use model on some benchmarks, but it is only available via an SDK behind a request form, so claims could not be verified hands-on.
ByteDance's OmniHuman image-to-avatar model goes public via Dreamina
ByteDance's impressive OmniHuman model, which turns a single image plus audio into a realistic talking avatar video, became publicly usable through the Dreamina (CapCut) website. The results land squarely in uncanny-valley territory, as Alex demonstrated with his own avatar thread.
Devin 2.0 launches with new IDE experience and $20/month entry price
Breaking during the show: Cognition Labs launched Devin 2.0, the second version of its AI software engineer, with a new IDE experience. Crucially, pricing now starts at $20/month, down from the original $500/month tier, making the agent far more accessible.
Claude gains Research mode and Google Workspace integration
Anthropic shipped a Research capability for Claude, letting it conduct multi-step research across the web, alongside a Google Workspace integration that connects Claude to email, calendar and docs context.
Veo 2 video generation hits GA in the API and Gemini App
Google made Veo 2 video generation generally available for developers and rolled it out in the Gemini App. The GA release brings Google's flagship text-to-video model out of preview and into production use.
W&B Weave Playground adds GPT-4.1 family and o3/o4-mini support
The Weights & Biases Weave Playground shipped full support for the new GPT-4.1 family and the o3/o4-mini models, letting developers evaluate and compare the week's new models for their own applications.
Google announces official support for the Model Context Protocol (MCP)
Demis Hassabis announced that Google will officially support Anthropic's Model Context Protocol (MCP) in its models and SDKs. This was a major signal of MCP becoming the industry standard for connecting AI models to tools and data.
OpenAI gives ChatGPT enhanced memory that can recall all your past chats
OpenAI rolled out enhanced memory for ChatGPT, allowing it to reference and recall all of a user's previous conversations rather than just saved memories. This makes ChatGPT significantly more personalized across sessions.
Google NotebookLM can now discover related sources for you
Google's NotebookLM added a source discovery feature that finds and suggests related sources for a notebook, instead of relying solely on user-uploaded documents. It extends NotebookLM further into research-assistant territory.
OpenAI ships new EMO "Monday" voice in ChatGPT
OpenAI added a new "Monday" voice to ChatGPT's voice mode, an EMO-flavored persona released around April 1st. It rounds out a week of OpenAI shipping across models, evals, and product.
Windsurf adds one-click deployments to Netlify
Windsurf shipped a deployments feature that lets users push apps straight to Netlify from the editor. A small but practical step toward end-to-end app building inside AI coding tools.
OpenAI's GPT Image generation lands in the API as gpt-image-1
OpenAI's powerful image generation capabilities, previously locked inside ChatGPT, are now available to developers via API under the official name gpt-image-1. This was the big one many developers were waiting for, opening up the viral image generation and editing capabilities for building AI art and image editing applications.
xAI finally launches the Grok 3 API tier
xAI made Grok 3 and Grok 3 Mini available via API, giving developers programmatic access to its frontier models for the first time. The Grok app also received updates the same week.
Hailuo Speech-02 TTS API: potentially SOTA emotional voice cloning
Hailuo (MiniMax) released the Speech-02 TTS API, which Alex called potentially state of the art for emotional control and voice cloning quality. It produces nuanced, realistic synthetic voices and was the standout voice release of the week.
Dex Horthy publishes 12-Factor Agents, a guide to production-ready agents
HumanLayer founder Dex Horthy published 12-Factor Agents, an open GitHub repo and essay distilling common patterns and pitfalls for building reliable, production-ready AI agents. Drawing on his experience building agent SDKs, it argues that serious teams end up writing large parts from scratch and lays out principles for robust agent design, discussed in depth on the show.
OpenAI debuts Codex CLI, an open source terminal coding agent
OpenAI released Codex CLI, an open source coding tool for the terminal. It ships with hardened security, using Apple Seatbelt on macOS to limit execution to the current directory plus temp files.
Cloudflare releases a new Agents SDK for building stateful AI agents
Cloudflare shipped a new Agents SDK for building and deploying AI agents on its edge platform. It joins the week's wave of agent infrastructure announcements alongside Google's A2A and broad MCP adoption.
GitMCP turns any GitHub repo into an MCP server instantly
Creators Liad Yosef and Ido Salomon launched GitMCP, a free tool that turns any GitHub repository into an MCP server by simply swapping the domain (gitmcp.io/user/repo). It lets AI assistants ground themselves in a repo's docs and code, and the creators joined the show to demo it.
ByteDance publishes Seed-Thinking-v1.5 reasoning model tech report
ByteDance's Seed team published Seed-Thinking-v1.5, a new reasoning model announced via a technical report on GitHub. It was mentioned among the week's open-source LLM news, though weights were not released at the time.
Test-Time Training paper one-shots minute-long videos with consistent characters
Researchers published 'One-Minute Video Generation with Test-Time Training', adding TTT layers to a pre-trained transformer to one-shot generate minute-long videos with remarkable character and scene consistency. The Tom & Jerry style demos showed the most impressive long-form AI video consistency to date.
Meta's MoCha generates movie-grade talking AI characters from speech and text
Meta GenAI researchers published MoCha, a model that generates stunningly realistic, movie-grade talking characters directly from speech plus text. Co-author Cong Wei joined the show to discuss the work, which points at AI actors entering Hollywood-quality territory.
OpenAI open sources the MRCR long-context benchmark dataset
OpenAI open sourced MRCR, a benchmark dataset for evaluating long-context, complex retrieval tasks, building on Gemini research from Google and publishing the dataset on Hugging Face.
CoreWeave hits 800 tok/s on Llama 405B with NVIDIA GB200 Blackwell
CoreWeave announced record-breaking AI inference benchmarks using NVIDIA's new GB200 Grace Blackwell superchips: 800 tokens/sec on Llama 3.1 405B, plus 33,000 tokens/sec on Llama 2 70B with H200s. It is a marker of how fast inference hardware is accelerating.
Gemini 2.5 Pro scores 24.4% on USAMO olympiad math, crushing the field
New evaluation results published this week showed Gemini 2.5 Pro scoring 24.4% on the USA Math Olympiad (USAMO), problems so hard that most top models score under 5%. The result showcases a step change in frontier reasoning ability on competition mathematics.
OpenAI releases PaperBench eval and open-sources Nano-Eval framework
OpenAI published PaperBench, a tough new evaluation that tests whether AI agents can replicate cutting-edge AI research papers, with more than 8,300 graded tasks and meta-evaluation of the LLM judge. The best model managed only a 21.0% replication score versus 41.4% for human PhDs. The code and the Nano-Eval framework were open sourced on GitHub alongside the paper.
OpenAI raises $40B at a $300B valuation
OpenAI closed a $40 billion funding round at a $300 billion valuation, one of the largest private raises ever. The show noted the raise rode the wave of native image generation in ChatGPT, with especially strong growth in India.
Google announces A2A, an open agent-to-agent communication protocol
Google announced the Agent2Agent (A2A) protocol at Cloud Next, an open spec for agents from different vendors to discover and communicate with each other. The spec was published on GitHub with a long list of launch partners, including Weights & Biases.
W&B launches observable.tools initiative and MCP observability RFC
Weights & Biases launched the observable.tools initiative and published an RFC (RFC-269) proposing observability standards for the Model Context Protocol, inviting community comment. W&B also announced it is a launch partner for Google's A2A protocol.
W&B launches Observable.tools initiative to add observability to MCP
Alex and Weights & Biases launched the Observable Tools initiative to bring observability to the Model Context Protocol (MCP) ecosystem, since external tool calls currently lose visibility for debugging and security. A concrete proposal using OpenTelemetry was posted to the MCP specification GitHub discussions for community feedback.
Don't wait a month — get every week's releases recapped in your inbox each Thursday.
