Multimodal

Alibaba opened Qwen 3.8 Max Preview on Alibaba Cloud, Qwen Chat, Qoder and the web, describing it as a 2.4T model headed for open weights. Early testers praised vision results but disputed coding claims.

Thinking Machines released Inkling with Apache 2.0 weights, 975B parameters, 41B active parameters, text/image/audio support, and up to 1M context. vLLM, SGLang, Modal, Databricks, and Vercel added day-zero support.

Meta said a model scored 30/30 on the APhO theoretical exam. Team posts described data curation, training and live participation, while public posts questioned which model was evaluated.

Goodside and other testers shared chat links where GPT-5.6 Sol, and sometimes Claude Fable 5, hallucinated hidden messages in noise images or meaningless scribbles. Higher effort settings sometimes did better, but failures reproduced.

LingBot 2.0 released code and weights for a real-time world model and robot-action models. The VLA maps 20 robot body configurations into a 55D action format and filters 90,000 raw robot hours to 50,000 training hours.

Riley Goodside tested random-noise and scribble images with no hidden message. GPT-5.6 Sol often produced invented text, while Claude Fable 5 more often refused or identified the image as non-writing.

Meta launched Muse Spark 1.1 in Meta AI and the Meta Model API public preview for coding, tool use, computer use, and multimodal reasoning. Early eval posts ranked it highly while system-card threads flagged safety details.

Goodfire introduced Block-Sparse Featurizers, which model activation concepts as multidimensional blocks instead of single SAE directions. The examples cover DINOv3, SDXL, and InceptionV1 activations.

Meta launched Muse Image in its apps and previewed Muse Video from the same media-generation family. Meta says Muse Image can reason, search, write code, self-refine, and use test-time compute before generating images.

OpenRouter tested 1,730 visual-reasoning questions across five models and found low-detail images often reduced accuracy while increasing reasoning-token spend. Caps on reasoning effort had the biggest billing impact.

X-Humanoid unveiled TG-VLA as a full-size whole-body VLA framework for humanoids, built around HEX, HAF-VLA, and DSRL-DCT. The company claims DSRL-DCT reached 100% success in mobile-manipulation tasks by freezing the VLA and learning a smaller noise-selection policy.

Gemini Omni Flash ranked #1 on Video Arena at 1404 Elo, 101 points above Seedance 2.0 Mini, and ComfyUI posted a text-prompt video-edit workflow. Google noted the leaderboard is third-party, leaving benchmark provenance as the main caveat.

Google shipped Nano Banana 2 Lite for image generation and Gemini Omni Flash for conversational video generation and editing in the Gemini API and AI Studio. The release sets image generation at about 4 seconds and $0.034 per 1K image, while Omni Flash adds multi-turn video edits at $0.10 per second.

Chandra's developer said Mistral OCR 4 launch numbers for both Chandra and OCR 4 could not be reproduced with public code, and published scripts to show the gaps. The dispute matters because Mistral OCR 4 launched on leaderboard claims, and benchmark settings now directly affect model selection.

Perceptron launched a video_frames input for Mk1 that accepts pre-decoded frames with timestamps instead of forcing clip re-encoding. The change matters for edge and sparse-footage pipelines because 10 minutes of 1080p video can start returning tokens roughly ten times faster.

A day after Seedance 2.0's 4K rollout story, partners began shipping the cheaper Seedance 2.0 Mini across Venice, ComfyUI, and Pika MCP. The 15-second 720p variant with native audio gives video workflows a lower-cost path than the flagship model.

Seedance 2.0 rolled out native 4K generation while Seedance 2.0 Mini landed on fal, Replicate, Pika MCP, and ComfyUI. That matters because engineers can now reach the same video model family through APIs, MCP workflows, and local graph tooling instead of a single web surface.

OpenRouter released a dedicated Image API that normalizes request shapes across 30-plus models from eight providers. Agents can inspect limits, passthrough options, streaming, and exact per-call cost without hardcoding vendor quirks.

Baidu released Unlimited OCR as an open-source long-document OCR model with 3B total parameters and 500M active at inference. Early ParseBench testing says it is strong on tables and reading order but weaker on semantic formatting and charts, giving teams a new open-weight OCR option with clear tradeoffs.

Mistral OCR 4 adds layout-aware extraction with bounding boxes, block typing, and inline confidence across 170 languages. Use it through the API or self-hosted deployments when document pipelines need structure, citations, redaction, and chunking.

Perceptron’s Files API lets developers upload an image or video once and reference it by ID across later requests instead of resending base64 or URLs. That simplifies repeated multimodal workflows and cuts transfer overhead for video-heavy pipelines.

Google put the Interactions API into GA as the new default for Gemini, adding background execution, managed agents, remote sandboxes, and multimodal tools. Builders now get one stateful interface for models, long-running jobs, and future Gemini Omni support.

lift-pdf released an open-source 9B model for schema-constrained document extraction, with code, pip install, playground access, and a 90.2% score on the team's 225-document bench. It matters because the model claims near-Gemini 3.5 Flash accuracy at 9.5s p50, though coverage is still skewed toward Latin-language docs and commercial-use limits remain.

Moonshot rolled out HighSpeed for Kimi K2.7 Code, claiming about 180 tok/s on coding tasks, up to 260 tok/s on shorter contexts, and roughly 6x speedups. Watch the tight capacity limits and mixed benchmark results, and budget for the 2x pricing if you want the faster mode.

ElevenLabs launched Music v2 on ElevenAPI with track generation, reference matching, inpainting, and multilingual output. It gives developers a priced API for commercial music creation and section-level editing.

Zyphra released ZONOS2 under Apache 2.0 with 8B total parameters, 900M active, zero-shot voice cloning, 44.1 kHz DAC audio, and ZTTS1-Eval. The release includes open weights, inference code, and eval code, so teams can run real-time multilingual TTS without a hosted-only stack.

MiniMax published M3 weights on Hugging Face with 428B total parameters, 23B active parameters, 1M context, and multimodal support. Unsloth quickly added local GGUF builds, so teams can try 2-bit runs at 138GB RAM or VRAM and 3-bit at 165GB.

Google released Gemini 3.5 Live Translate for low-latency speech translation across 70+ languages in the Gemini Live API, AI Studio, and Google Translate. The same model is also heading to Google Meet in private preview for Workspace customers.

Posts from WWDC say Apple Intelligence now combines Apple Foundation and Gemini models, and Siri gains visual, on-screen, and app-level actions. Watch for the beta rollout later this year; multiple posts say it will not ship in the EU at launch.

Google released Gemma 4 12B, an Apache 2.0 encoder-free multimodal model with native audio and vision for 16GB-class laptops. Day-zero support in llama.cpp, vLLM, Ollama, MLX, and SGLang should make local agents and on-device apps easier to deploy immediately.

Two days after Qwen 3.7 Plus launched, Hyper, OpenCode, Kilo, and Vals shipped support or rankings around the 1M-context multimodal model. The rapid pickup shows Alibaba’s new model landing quickly in coding-agent tools and public eval stacks outside its own platform.

Microsoft introduced MAI-Thinking-1, MAI-Code-1-Flash, and five other MAI models across code, image, voice, and speech. The launch puts Microsoft back into the frontier-model race and starts landing pieces of the stack in Copilot and partner runtimes.

H Company released Holo 3.1, a local computer-use VLM family with function calling and AndroidWorld gains up to 79.3% on the 35B model. The update pushes computer-use agents toward local and mobile deployment instead of cloud-only runtimes.

NVIDIA released Cosmos 3 as an open omnimodel family with 16B and 64B variants, plus code, datasets, and a coalition around physical AI. The release matters because it ships with serving support and top open-weight image and video rankings, so teams can use it beyond a research teaser.

Alibaba released Qwen 3.7 Plus as a multimodal agent model for GUI, CLI, coding, and browser tasks. It ships with browser demos and immediate Cline support, giving teams another frontier-style agent model to compare against M3 and closed-source tools.

MiniMax shipped M3 with a 1M-token context window, native multimodal input, and frontier coding claims across SWE-Bench Pro, Terminal Bench, and MCP Atlas. It also appeared on OpenRouter, Ollama Cloud, Venice, Hermes, Cline, Together, and Arena on day one.

Grok Imagine Video 1.5 moved from arena ranking to usable APIs, with xAI docs live and third-party access on fal and Venice. That matters because developers can now script against the model through standard providers, though early #1 arena claims are already being challenged by side-by-side testers.

A day after launch, Nous made Step 3.7 Flash free for 30 days to Hermes users through Nous Portal. The access window landed alongside fresh vLLM/NIM and MLX-VLM support, making the model easier to test in both local and production stacks.

Grok Imagine Video 1.5 Preview took the top 720p Image-to-Video Arena slot with a reported 52-point gain over the previous Grok video model. xAI docs and shared console pricing put the model at $0.08 for 480p and $0.14 for 720p, giving developers a concrete new API option for video generation.

Step 3.7 Flash landed immediately across Kilo, Modal, SGLang, Hermes-linked tooling, and DesignArena as the model’s 198B MoE, 256K-context release spread through the stack. The breadth of day-one support gives engineers multiple ways to serve, benchmark, and wire the new open-weight multimodal model into agents.

Google moved Nano Banana 2 and Nano Banana Pro to GA in AI Studio and the Gemini Enterprise Agent Platform. Nano Banana 2 also takes video as input, giving image pipelines published per-image pricing and a production API.

Google expanded SynthID with new model partners and pushed verification into Search, Chrome, and Pixel video provenance flows. That matters because AI-content authentication is moving from isolated model outputs into mainstream browser and distribution surfaces.

Cohere open-sourced Command A+, a 218B MoE multimodal model with 25B active parameters, 48-language support, and deployment starting at two H100s. Artificial Analysis put it at 37 on its Intelligence Index and 281 tok/s, and vLLM plus Transformers added support.

Google launched Gemini Omni Flash as its first shipping any-input-to-video model, with character consistency, physics-aware scenes, and conversational video editing. Use it in Gemini, Flow, and YouTube surfaces first, and wait for API access if you need programmatic integration.

Leak videos and tester reports pointed to a larger Gemini desktop app with Stream to Cursor, Spark local-file access, Live, and Omni ahead of I/O. Independent testers also reported faster 3.2 and 3.5 Flash checkpoints, but Google had not announced the features publicly.

Alibaba put Qwen3.7 Max Preview and Qwen3.7 Plus Preview live on Arena and the Qwen site, with Arena placing Max Preview #13 overall and #10 for coding. That gives engineers an early read on the next Qwen generation before any broader API or open-weight release.

Perceptron launched Mk1, a multimodal model for video and embodied reasoning with native 2 FPS video, 32K context, and structured spatial outputs. OpenRouter access and the low input price make it usable for deployment, not just demos.

Google unveiled Gemini Intelligence at the Android Show with cross-app task automation, Gemini in Chrome, Rambler voice cleanup, custom widgets, and AppFunctions. The rollout moves Gemini into core Android workflows on Pixel and Galaxy devices this summer.

Hugging Face released Diffusers 0.38.0 with new audio and image pipelines, Flash Attention 4, FlashPack loading, and Ring Anything for context parallelism. Use the new profiling guidance to tune diffusion performance as you adopt the added model coverage.

Thinking Machines previewed interaction models that process audio, video, and text in 200 ms micro-turns, letting the system listen, speak, and react at the same time. The demos matter because the interaction loop is trained into the model instead of stitched together from separate speech and tool layers.