Multimodal

Simon Willison tested Alibaba's Qwen 3.8 27B and found the default xhigh reasoning effort causes absurd overthinking. A simple circle prompt triggered minutes of animated SVG generation; a pelican-on-a-bike SVG burned 22,276 reasoning tokens over 21 minutes. Turning reasoning off cut the same task to just over two minutes. He recommends starting with low or no reasoning. The model also nailed bounding-box detection on a pelican photo with near-perfect accuracy.

A woman, Jane Doe 4, joined a class-action lawsuit against xAI, alleging her stepfather used Grok to create over 7,000 explicit images from a photo taken when she was 11. Her stepfather died by suicide two days after a law enforcement raid uncovered the images. Three Tennessee teenagers had previously sued xAI, claiming Grok lacked basic safeguards to prevent generating explicit imagery of real people, including minors. Earlier this year, X was flooded with millions of Grok-generated sexualized images. TechCrunch has reached out to xAI; the post does not include a response.

Fudan's MOSS-VL makes real-time interaction—perceiving while speaking—a first-class capability. Gated cross-attention keeps visual tokens outside the decoded sequence, giving it a 2.8× to 5.1× time-to-first-token advantage over same-backbone Qwen3-VL-8B. MOSS-VL-Realtime tops three of four streaming benchmarks, hitting 66.0 vs. 37.5 on OmniMMI Proactive Alerting. The offline variant leads temporal-reasoning video sets at comparable scale. All five checkpoints, the training curriculum, and inference code are open.

DeepSeek V4's API rejects image input, but Liquid AI's new LFM2.5-VL-3B can serve as a perception layer. The author tested it zero-shot on garage camera data — 94.3% accuracy for door open/close, 1.5s locally. Perception runs on-device, images never leave, only structured text hits the cloud for reasoning. The latency, privacy, and cost advantages of this split architecture hold even if DeepSeek adds native vision later. One gap: extracting a reliable confidence signal from natural-language output — the post doesn't detail a calibration method.

Qwen delivered on its open-source promise with the Qwen3.8 series. Qwen3.8-27B is a natively multimodal dense model that beats Qwen3.7-Plus at only 27B parameters, supports 262K context natively and up to 1M tokens via YaRN, under Apache 2.0. Open weights for the Max-tier Qwen3.8-2.4T-A95B are also available. The post doesn't cover training data, inference cost, or release timeline details.

The big one today: researchers extracted hidden reasoning chains from Anthropic, OpenAI, and Google models at scale. The trick is absurdly simple—take Opus 4.8's encrypted CoT and feed it to Haiku 4.5, which decodes it verbatim. All three API families were broken, and decoding 10K trajectories costs about $720. A separate paper shows you can even reverse-engineer reasoning from public outputs alone using a 1.5B-param model. Separately, Coze CLI was caught silently scanning local Codex and Claude Code directories and injecting its own skills into workflows. On the engineering side, the group discussed how prompt debt now rivals traditional code debt—old rules pile up, evals lag behind model iterations, and nobody dares delete anything.

Google DeepMind released SL2T, a multilingual sign language-to-text model, bringing sign language AI into consumer products for the first time. On Pixel 11, Gboard and Live Transcribe support American Sign Language (ASL) to English dictation, with more devices and languages to follow.

Sundar Pichai announced on X that the standalone Gemini app surpassed 1 billion monthly active users—Google's 14th product to hit that mark. The figure excludes AI Mode in Search and other channels. ChatGPT reached 1B MAU in June; Gemini is now keeping pace. Usage stats: 63% of users have tried the voice feature, the app generates over 150 million images daily, and iOS has more than 100 million active users. The post doesn't disclose paid user share or revenue.

Meta released Muse Glimmer on Monday, an open-weight AI model nearly identical to its paid, closed-source Muse Spark launched in July—capable of generating code, text, and images. Mark Zuckerberg also published a 14-page essay arguing superintelligence should not be concentrated in a few companies, and announced a $1 billion fund for communities hosting its data centers. Muse Glimmer is open-weight, not fully open-source; the underlying code isn't fully public. Meta also teased a more powerful model codenamed Watermelon but didn't disclose whether it will be open or closed.

Meta released Muse Glimmer, a 30B multimodal model built for local agentic workflows. SGLang ships Day-0 support with dedicated optimizations: on a single RTX 5090 with NVFP4 quantization and DFlash speculative decoding, per-user decode hits 236 tok/s and total throughput reaches 1,452 tok/s. The model uses a hybrid of sliding-window and full-sequence attention with a 128k+ context window. Apple Silicon is supported via the MLX backend, though speculative decoding isn't available there yet.

Meta released Muse Glimmer weights under Apache 2.0. It's a 30B model built for always-on local agent workflows, small enough to run on a Mac or PC with a single consumer GPU. 4-bit quantization shrinks it below 20 GB, leaving room for KV cache and the vision encoder within a 24 GB or 32 GB envelope. Training used logit distillation from a larger Muse Spark teacher, followed by mid-training on long-context agent data and post-training with SFT, on-policy distillation, and RL. Meta's benchmarks show it outperforming Gemma4-31B and Qwen3.6-27B on agentic, coding, multimodal, and safety evals. The post doesn't disclose specific latency numbers, only that inference optimizations were applied to keep it responsive.

Qwen-Image-3.0 is pitched as production-ready: 4.5K-token prompts, 100%+ text accuracy with no broken logos, and native support for 12 languages. High-res generation starts at $0.03. The post only provides a headline and links—no model architecture, inference speed, or benchmarks are disclosed, so I'd hold off on the accuracy claim until third-party tests appear.

Mistral AI launched Shieldstral on Product Hunt this week—a 3B open-weight multimodal guardrail. You define safety policies in natural language at inference time, and it evaluates text, images, or both with a single token output. It runs locally on one 16GB GPU. The Product Hunt page shows the basics and screenshots, but doesn't disclose latency, accuracy benchmarks, or comparisons with other guardrails like Llama Guard. Treat it as a lightweight, self-hostable compliance component for now; real-world performance needs community benchmarks.

Simon Willison fed a 2022 tweet and two concept images to Claude Fable 5 and let it build a playable browser 3D game with zero further input. The model chose Three.js, called OpenAI's gpt-image-2 for textures, and added mechanics like a patrol dog with scent tracking. The whole project was done on mobile, deployed via GitHub Pages. The gameplay is basic, but the zero-intervention workflow is the real story.

Qwen Image 3.0 Pro handles up to 4,500 input tokens and generates dense layouts—newspapers, storyboards, menus—in one pass. It reliably renders text down to 10px across 12 languages and 20+ fonts, and reproduces micro-expressions, pores, and hair strands at near-photographic quality. Output pricing is $0.04 per 1K image and $0.075 per 2K image, but the rate limit is just 1 request per minute, so high-throughput use cases are off the table for now.

Mistral introduced Shieldstral, a 3B-parameter open-weights model built to moderate both text and image content. It can run locally or on edge devices, checking user inputs and model outputs for policy violations. The post doesn't disclose benchmark scores, latency figures, or pricing—only that it's positioned as a safety filter. I'd wait for third-party evals, but the 3B size is genuinely lightweight for self-hosted moderation.

SenseTime open-sourced SenseNova U1, a model that handles reasoning and image generation in a single pipeline. It can turn a prompt into a structured slide deck or generate step-by-step illustrated content, like a six-step dragon drawing tutorial. Available on HuggingFace, GitHub, and SenseNova Studio. The post doesn't disclose parameter count, training data, or benchmarks.

Qwen 3.8 Max launched with Terminal Bench 2.1 score 86.6 and PaperBench 93.0, beating Fable 5's 88.8. A 500-yuan token plan burned out in one day; the model lands between Luna and Terra, with price as the main draw. Open weights for both Qwen 3.8 Max and Qwen 3.8-27B drop next week. DS V4 Flash hit 8T tokens consumed in a single day, topping weekly charts—the group sees tokens becoming a commodity. On tools: an M5Stick voice dongle turns a keychain into an agent remote, LoopX keeps agent state across 200+ hours, and reverse-skill injects reverse-engineering toolchain knowledge into coding agents. The wildest methodology story: an agent autonomously downloaded a local Qwen mid-translation task and auto-installed Whisper when its API key ran out of funds.

Alibaba Qwen announced Qwen3.8-Max, a 2.4T-parameter model, and Qwen3.8-27B, both promised as open weights. Max claims 10+ days of autonomous coding, a 125-hour self-directed research loop beating the original paper by 2.71 points, and a 4.16x return in a 365-day e-commerce sim. API pricing is $2/M input, $6/M output. I'd hold the champagne: the post doesn't include standard academic benchmarks, and the exact open-weight date and license aren't specified.

As of today, tech companies in the EU must label AI-generated content and disclose when users interact with a machine. Fines reach €15 million or 3% of global annual turnover. The rules cover chatbots, deepfakes, and AI-generated audio, video, and text. The EU also published official label icons companies can use instead of designing their own.