#Model release

AIBlack Forest Labs has released flux-3-action-base, an open-weights 7B world action model that takes camera frames, robot state, and a text instruction to output the next action chunk. The release is an adaptation component rather than a complete robot policy, and new embodiments require their own action heads. The source says the weights are paired with shared video VAE and Qwen3-VL-4B-Instruct text encoders and is governed by the FLUX Kommunity License v.1.0.

Why it matters: The source separates the adaptation base from full robot policies and states the shared encoders and new-embodiment requirements, which clarifies what developers must still build for their robots.

AIMETR's preliminary evaluation concludes that Claude Opus 5.5 likely gives slightly higher AI R&D productivity uplift than Fable 5.1 but is unlikely to fully automate AI R&D. The evaluation used five capability tasks over 10 business days of API access, and METR says Anthropic reviewed and edited the summary before sign-off.

Why it matters: The report separates two claims about AI R&D acceleration and discloses that Anthropic reviewed the summary, which helps readers weigh its independence and evidence.

AIGoogle made the Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS models generally available, along with the Gemini API Voices endpoint. Flash TTS is positioned for studio-grade voice fidelity and long-form multi-turn stability, while Flash-Lite TTS targets high-throughput, real-time voice agents and replaces gemini-3.1-flash-tts-preview. The update adds voice design, voice replication with consent verification, and access to 150+ prebuilt and custom voices.

AIAnthropic has launched Claude Opus 5.5, the first model in its new Claude 5.5 family. The company says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5.

Why it matters: The release note gives a direct comparison to Claude Fable 5.1 and a 40% running cost reduction versus Opus 5, which helps readers weigh the tradeoff.

AITogether AI's canary rollouts shift production traffic between two model deployments on the same endpoint in staged percentages, with optional metric gates between steps. Operators can choose canary, blue-green, or rolling strategies, and a rollout starts only when explicitly launched; it can be paused, canceled, or reversed. The platform scales the target before moving traffic and waits for routing to converge before draining the source.

AIXiaomi MiMo released MiMo-V2.6-Distill-Qwen-9B, a 9B agentic model made by supervised fine-tuning Qwen3.5-9B on MiMo-generated data, as an open starting point for agentic reinforcement learning research. It scored 61.1 on SWE Verified, versus 60.0 for Qwen3.5-9B, and 44.6 on SWE Pro, versus 32.0. The checkpoint is served with SGLang and a MiMo chat template, and its SFT data totals 77.4B tokens.

AIApple has released LensVLM-9B on Hugging Face, a 9B-parameter Vision Language Model that scans compressed images of text and selectively expands relevant pages to their uncompressed form. The repository provides a demo script and supports compression settings of 5x, 10x, and 15x. Model files are under the Apple Machine Learning Research Model License, and the accompanying source code is distributed separately under the Apple Sample Code License.

AIXiaomi released MiMo-V2.6-Flash-RL, an efficiency-balanced checkpoint in its MiMo-V2.6 series, on Hugging Face. The model is a sparse MoE with 309B total and 15B activated parameters, supports text, image, video, and audio input, and offers a 1M-token context. The technical report says it was trained with a single mixed reinforcement learning run across coding, agent, visual, and cybersecurity tasks.

Why it matters: The report pairs its benchmark tables with the RL training method, which helps readers judge how the checkpoint's scores relate to its training approach.

AIXiaomi MiMo released MiMo-V2.6-Pro-RL on Hugging Face, a sparse MoE model with 1.02T total and 42B activated parameters and a 1M-token context. The technical report says it accepts text, image, video, and audio, and was trained with a single mixed reinforcement learning run across coding, agent, visual, and cybersecurity tasks.

Why it matters: The report pairs a 1.02T-parameter MoE model with an RL-based self-improvement method, useful for judging how reinforcement learning is scaled in frontier open models.

AIxAI released Grok 4.7, which it calls its most capable model for coding and knowledge work, built on a larger base model than Grok 4.6 and trained with a longer reinforcement learning run. It is priced from $2 per million input tokens and $6 per million output tokens, the same as Grok 4.6, and is available in Cursor, Grok Build, and the Grok API. xAI reports gains on CursorBench 4.0 (46.3%) and AA Briefcase v1.1 (1,657) over Grok 4.6, and says it posts the strongest safety results it has tested on refusals and jailbreak resistance.

Why it matters: The release pairs a new base model with benchmark tables against named rivals and pricing, letting readers compare its coding and office-work gains against Grok 4.6 and frontier models.

AIQwen has open-sourced Qwen-Image-2.1, a unified text-to-image generation and image editing model with 7B visual generation parameters. The Hugging Face page for Qwen-Image-2.1-PE-I2I is a fine-tuned Qwen3.5-VL 9B prompt rewriter that turns vague editing instructions and input images into precise editing prompts, supporting up to 10 reference images.

Why it matters: The model card documents usage with transformers and diffusers, letting readers see how the editing prompt rewriter connects to the generation pipeline.

AIQwen has open-sourced Qwen-Image-2.1, a unified text-to-image generation and image editing model with a 7B-parameter visual generation component. The release also includes Qwen-Image-2.1-PE-T2I, a fine-tuned Qwen3.5-VL 9B model that rewrites brief image requests in any language into detailed English prompts with a recommended aspect ratio.

Why it matters: The release pairs a 7B visual generation component with a separate prompt rewriting model, showing how a brief image request becomes a detailed English prompt before rendering.

AILiquid AI released LFM2.5-VL-3B-DSpark, a speculative-decoding draft model for its LFM2.5-VL-3B vision-language model. The source reports decoding up to 2.66× faster on a single H100 with SGLang, up to 3.13× on Apple M5 Max with MLX-VLM, and up to 2.14× on Apple M3 Ultra with llama.cpp, with output unchanged under greedy decoding.

AIxAI released Grok Voice Transcribe 2.0, a speech-to-text model that is twice as accurate as Grok Voice Transcribe 1.0 at the same price, and ranks first for accuracy among 32 streaming models on the Artificial Analysis leaderboard. Batch transcription costs $0.10 per hour of audio and streaming $0.20 per hour, with diarization, timestamps, and key terms included. Existing Speech-to-Text API integrations gain the improvement with no code changes, and developers must pin grok-voice-transcribe-1.0 to stay on the older model during the transition.

AIinclusionAI has released Ming-Image-0.1-Design-Layer on Hugging Face, a model that decomposes a flattened design image into a requested number of RGBA layers using an image and a layer plan. The model runs at 1024 resolution (512 for faster processing) with 12 sampling steps, a CFG scale of 2.0, and BF16 precision on one CUDA GPU with 80 GiB VRAM. It is released under the MIT License.

AIinclusionAI has released Ming-Image-0.1-Design, a 6B text-to-image model for UI, infographics, and posters that outputs RGBA images with transparent backgrounds. The model is available on Hugging Face and ModelScope under the MIT License. It runs at 2048 x 2048 with 12 sampling steps and a CFG scale of 1.0, validated on one CUDA GPU with 80 GiB VRAM.

AIinclusionAI has published Realtime-Venus on Hugging Face with two 9B checkpoints: Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for audio-only conversation. Both are built on MiniCPM-o 4.5 with a Qwen3-8B backbone and support full-duplex dialogue, proactive responses, and training-free long-video memory. The asynchronous Realtime-Venus-Harness runtime is hosted in a separate GitHub repository.

AITencent's WeVisDoc-4B, fine-tuned from Qwen3-VL-4B-Instruct, converts page images into structured Markdown with LaTeX formulas and HTML tables. It scores 95.38 Overall on OmniDocBench v1.6 and a mean Overall of 75.54 across three PureDocBench tracks, ranking first among compared end-to-end parsers in all four reported settings. The model is available on Hugging Face and runs through vLLM, which requires version 0.11.1 or later.

AITencent's WeVisDoc-4B, fine-tuned from Qwen3-VL-4B-Instruct, scores 95.38 Overall on OmniDocBench v1.6 and 75.54 mean Overall across three PureDocBench tracks. The end-to-end parser converts page images into structured Markdown with LaTeX formulas and HTML tables, and the 2B variant is also available. The repository provides vLLM serving scripts with a 32768-token default context and a Python client for batch processing.

AITogether AI says teams can migrate from closed to open-source models in weeks to months, not years, using a managed service. The guide recommends defining the use case, shortlisting candidates with relevant benchmarks, and replaying existing traffic against them to test accuracy, cost, and latency against current targets.

AIGoogle DeepMind introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two live dialogue models for voice agents. Extended Thinking scores 82.6 on Artificial Analysis' Speech to Speech Quality Index, 68.6% on τ-Voice, and 97.7% on Big Bench Audio. Gemini 3.8 Live is rolling out now in the Gemini API, Google AI Studio, and Search Live, with enterprise access in private preview.

Why it matters: The release covers a voice model's benchmark results and availability across developer, enterprise, and consumer products, useful for judging voice agent options.

AINVIDIA's RT-DETR Hand Detection v1.0 detects and localizes left and right hands in RGB images, outputting 2D bounding boxes with per-hand confidence scores in a single pass. The model, built on RT-DETRv2-S with HGNetv2-S backbone and about 20M parameters, is intended as a region-of-interest stage for downstream 3D hand pose estimation and is exported to ONNX. The source describes it as for demonstration purposes rather than production use, runs on NVIDIA Lovelace GPUs under Linux, and is licensed under the NVIDIA Software and Model Evaluation License.

AILangChain is using Baseten Loops, a managed fine-tuning service, to train custom models for LangSmith Engine, its in-platform agent that debugs and improves AI agents. The article says LangChain fine-tunes large open-weight models on agent traces and trains smaller open-weight models such as Qwen for tasks like failure-mode categorization. Baseten Loops supports supervised fine-tuning, reinforcement learning, and long-context workloads, and lets checkpoints be evaluated and deployed directly to inference.

AIGoogle has made two audio-to-audio models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, generally available through the Live API. Gemini 3.8 Live, model ID gemini-3.8-live, is the default for low-latency voice agents, with interleaved reasoning and asynchronous function calling. Gemini 3.8 Live Extended Thinking, model ID gemini-3.8-live-extended-thinking, supports background reasoning during live audio and is recommended when more reasoning is needed.

Why it matters: The changelog names two model IDs and their intended use, showing how Live API developers can choose between low-latency voice and higher background reasoning.

AINVIDIA Research released FoundationStereo-small, a zero-shot stereo depth model that takes an RGB stereo pair and outputs a disparity map, on Hugging Face. The model has about 6.3×10^7 parameters and ships as ONNX files at fixed 576x960 and 320x736 resolutions, with TensorRT and ONNX runtime support. It is licensed under the NVIDIA Open Model License and is ready for commercial use.

AINVIDIA released FoundationPose, a transformer-based model for 6-DoF object pose estimation and tracking that works on novel objects at test time without fine-tuning, given a CAD model. It takes RGB and depth images, a 2D bounding box, a CAD model, and camera intrinsics as inputs, and is licensed under the NVIDIA Open Model License for commercial use. The model is trained on synthetic data from Objaverse and Google Scanned Objects, with evaluation on LINEMOD and YCB-Video.

AIGoogle DeepMind released EmbeddingGemma 2, an open model under Apache 2.0 that maps text, images, video, and audio into one shared 768-dimensional vector space. The model has 740M total parameters and supports 8,192-token context, with Matryoshka truncation to 128d, 256d, and 512d. The source reports 14% better code-task performance than EmbeddingGemma 1 and says it is designed for consumer hardware such as phones and laptops.

Why it matters: The release combines text, image, video, and audio retrieval in one 768-dimensional space at 740M parameters, a useful reference for on-device multimodal search design.

AITencent released Simple-Attention-Sparsification (SAS) gate checkpoints for Qwen3-4B, Qwen3-8B, and Qwen3-14B, which learn to rank and select KV blocks using continuous gates optimized with the language-modeling loss. The router-only packages, 64 MiB to 81 MiB each with 33.0M to 42.0M gate parameters, require the frozen Qwen3 base model and the seer_attn backend in a forked sglang-blocksparse build. The default sparse decode budget is 2,048 tokens, and the checkpoints can be evaluated at 1,024, 2,048, or 4,096 budgets without retraining.

AIQwen has open-sourced Qwen-Image-2.1, a unified text-to-image generation and image editing model with 7B parameters in its visual generation component. The model can generate regular or transparent RGBA images, supports up to 10 reference images for editing, and is licensed under the Qwen Research License Agreement.

Why it matters: The source specifies the 7B visual component, transparent RGBA output, and up to 10 reference images, which helps readers judge its fit for generation and editing workflows.

AIinclusionAI released Step-3.7-Flash-singprobe, an 8.13M-parameter probe that reuses Step-3.7-Flash hidden states to score query intent, response unsafety, and hallucination risk at every generated token. The probe adds less than 0.5% decode-time overhead and reports 0.9858 R-AUC and 0.9295 T-AUC on streaming safety benchmarks. It is supported through SGLang and vLLM integration branches and loads from Hugging Face by checkpoint ID.

AIinclusionAI has released Qwen3.8-27B-singprobe, a 10.1M-parameter intrinsic streaming guardrail that reuses Qwen3.8-27B hidden states to score query intent, response unsafety, and hallucination risk at every token. The probe adds less than 0.5% decode-time overhead and reports a 0.03% benign-response false-positive rate averaged across five datasets. It is supported through SGLang and vLLM integration branches, with training code available at inclusionAI/SingProbe.

AIinclusionAI has released Qwen3.5-397B-A17B-singprobe, an intrinsic streaming guardrail built on Qwen/Qwen3.5-397B-A17B that scores query intent, response unsafety, and hallucination risk at every generated token using the base model's hidden states. The probe has 8.13M parameters, taps layers 18, 38, and 58, and adds less than 0.5% decode-time overhead. Training code is available at inclusionAI/SingProbe, and the probe runs through SGLang or vLLM integration branches.

AIinclusionAI released SingProbe, a 5.8M-parameter intrinsic guardrail built on openai/gpt-oss-120b that scores query intent, response unsafety, and hallucination risk at every token. It reuses the base model's hidden states, adding less than 0.5% decode-time overhead, and reports a 0.06% benign-response false-positive rate. The probe is available on Hugging Face and supported through SGLang and vLLM integrations.

AIFireworks AI reports that DeepSeek-V4.1-Flash scores 74.34% pass@1 on DeepSWE at $0.430 per task, close to GPT-6-Astra's 74.12% at $6.524. On Terminal-Bench 2.1 it scores 86.5% against Astra's 87.5% at about 12x lower cost per task, while on HLE it trails Astra alone at 34.52% versus 50.40%. The post also reports that a combined oracle router reaches 54.80% on HLE, and that serverless and dedicated API access is available with US-hosted endpoints coming soon.

AIShanghai AI Lab's InternLM team released Intern-S2-397B, a multimodal foundation model for scientific intelligence and long-horizon agents. The model uses visual pre-training on raw scientific literature pages, multi-task reinforcement learning across more than 20 scientific domains, and agentic reinforcement learning in sandboxed environments.

AIDeepSeek released open weights for V4.1-Flash, which Baseten now offers through its Model APIs. The model has 552B total parameters, 8B active for prefill and 16B for decode, a 1M token context window, and text plus image input. Its Causal Encoder-Decoder design runs only the encoder during prefill and reuses a projected KV cache, and the source reports the global KV cache at a quarter of V4-Flash's memory.

Why it matters: The post explains how the CED architecture splits prefill and decode compute and cuts KV cache memory, which matters for coding agent costs.

AICognition is making Fusion available in Devin Desktop and CLI, a harness where a frontier lead model plans and reviews while a cheaper sidekick executes. Across listed coding benchmarks, Cognition reports Fusion cuts cost per task by about 11% to 46% versus the lead model alone, while the sidekick does the implementation work. The post recommends pairing Fable 5.1 with SWE-2, and argues price per task matters more than price per token.

AIShanghai Artificial Intelligence Laboratory has released Atria Dawn Preview, an agentic model built on the 744B-parameter MoE GLM-5.2 foundation model, with a 256K context window. The release page reports benchmark results across search, coding, tool use, productivity, and cybersecurity, and describes text-only setup for Codex and Claude Code.

Why it matters: The release page gives a full benchmark table against named rivals and setup steps for Codex and Claude Code, useful for anyone evaluating agentic models.

AIAi2 released AstaBrief-8B-SFT, an 8B intermediate supervised fine-tuning checkpoint built on Qwen3-8B that turns a research question and retrieved literature excerpts into a cited report. On the ScholarQA-CS2 test set of 100 computer science questions, it scored an average of 83.7 versus 77.3 for base Qwen3-8B, with citation recall at 71.3 versus 64.6. The model is licensed under Apache 2.0 for research and educational use.

AICognition introduces SWE-2, a coding model post-trained from Kimi K3 that scores 50.0% on FrontierCode 1.1 Main, within one point of Fable 5.1 while costing 64% less. The post attributes the gains to an RL algorithm that trains all reasoning-effort levels in one run, with cost penalties tuned to the base model's Pareto frontier. SWE-2 is available starting today in Devin Desktop and CLI, with rollout to Devin Web and Fusion.

Why it matters: The post explains how the cost penalty and length-weighted baseline are derived, which helps readers judge the tradeoffs in coding model post-training.