Every AI model release we saw in the last 30 days, open and closed, newest first.
| Date | Model | By | Open? | Size | Context | License | Price |
|---|---|---|---|---|---|---|---|
| 14 Oct | Grok Voice 3.0 xAI released Grok Voice 3.0, a speech-to-speech model for building voice agents; it is reportedly already handling Starlink and Tesla calls. | xAI | Closed | ||||
| 14 Oct | HiRO-ACEv2 HiRO-ACEv2 is an AI weather model that simulates extreme weather at 3 km resolution for a fraction of the usual cost, released on Hugging Face. | HiRO | Open | ||||
| 14 Oct | Ling-3.1-flash An open Hugging Face model with 560B total parameters, 25B active per token and a 1M-token context window. It follows Ling 3.0, which was 127B. | inclusionAI | Open | 560B total, 25B active | 1M | ||
| 14 Oct | Nemotron Bolt 3.5 A Reddit post says Nvidia's Nemotron Bolt 3.5 is a 74B-parameter, 9B-active MoE model quantized to 24GB using 2-bit quantization-aware training for speed. | NVIDIA | Open | 74B A9B MoE | |||
| 14 Oct | Orion-1 Artemis Security's Orion-1 is a model purpose-built for cyber defense that beats Claude Opus 5.5 and GPT-6 Sol on the company's benchmark, filling a gap as frontier models lock down cyber capabilities. | Artemis Security | Closed | ||||
| 14 Oct | GLM 5.3 BigGo Finance argues Z.ai's GLM 5.3 shows post-training techniques, rather than raw compute, have become China's key competitive advantage in AI. | Z.ai | Open | ||||
| 14 Oct | Maya-2 An X post announces Maya-2, billed as the world's best Indic voice AI model with fully open weights, claiming it outperforms ElevenLabs v3, Cartesia Sonic 3.5 and Bulbul v3 across Indian languages; free to builders soon. | Maya-2 | Open | open-weight | |||
| 14 Oct | Neuron AI German robotics unicorn NEURA Robotics introduced Neuron AI, a multimodal foundation model that gives robots perception and memory for physical AI and humanoid applications. | NEURA Robotics | Closed | ||||
| 13 Oct | Aziib Pixel Style A pixel-art LoRA for image models Krea2 and Anima, downloadable from Civitai with the trigger word 'aziib pixel style'. | Open | |||||
| 13 Oct | LFM2.5-24B-A2B LFM2.5-24B-A2B is an open-weight Mixture-of-Experts model from Liquid AI with 24B total and about 2B active parameters, designed to run fast, private agent tasks locally on a laptop. Weights are available on Hugging Face. | Liquid AI | Open | 24B (A2B active) | |||
| 13 Oct | Grok 4.7 xAI's Grok 4.7 model is now available on Amazon Bedrock. AWS shows how to build a tool-calling agent and evaluate it across three reasoning tasks at four reasoning effort levels, measuring correctness and token cost. | AWS | Closed | ||||
| 13 Oct | Kugel 3.5 Kugel 3.5 is a text-to-speech model its makers claim outperforms other models on the market. It was shared on Product Hunt; no pricing or availability details were given. | Kugel | |||||
| 13 Oct | Qwen2.5-1.5B-MoE-A0.9B An independent developer continued work on converting dense LLMs into Mixture-of-Experts models, releasing Qwen2.5-1.5B-MoE-A0.9B along with GGUF versions for local use. | Open | 1.5B total, 0.9B active | ||||
| 13 Oct | EmbeddingGemma 2 A project repurposes Google's EmbeddingGemma 2 embedding model into a speech-to-text system, with the write-up and demo shared on Hacker News. | Open | |||||
| 13 Oct | GLM-Legal Z.ai's GLM-Legal, built on GLM-5.3-Flash, is a legal AI model scoring 19.2% all-pass on Harvey's Legal Agent Benchmark, nearly 3x its base, at $0.15 per test, by application. | Z.ai | Closed | $0.15 per test | |||
| 13 Oct | Natura-1 Natura-1 is a 50B-parameter Mixture-of-Experts general foundation model for the physical world, pretrained on one million hours of egocentric data. It follows natural-language instructions zero-shot, driving dexterous hands. | 50B | |||||
| 13 Oct | NiCLIP Nature presents NiCLIP, a neuroimaging contrastive language-image pretraining model that predicts text from brain activation images. | ||||||
| 13 Oct | GreenWAM GreenWAM is a world foundation model that both predicts how a robot's actions change a scene and chooses those actions, aiming to work across different robot bodies. | ||||||
| 12 Oct | Gemini 4 Argon Google released Gemini 4 Argon, its next flagship model, with safety work, robotics applications, and a demo of optimizing a 200,000-line Python codebase overnight. | Closed | proprietary | ||||
| 12 Oct | Flux Deepgram is expanding its Flux voice model to 10 languages and targeting air-gapped laptops, letting voice AI run locally without cloud connectivity for sensitive environments. | Deepgram | Closed | ||||
| 12 Oct | JEV-35B JEV-35B is an Apache-2.0 text-classification model built on Qwen3.5-35B-A3B, with NVFP4 and GLM5.3-Flash variants also published. A GuruSearch recommendation and search demo is live, and it has 6.3k downloads. | AutoTrust | Open | 35B | apache-2.0 | ||
| 12 Oct | Muse Wonder / Muse Wonder Full Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full, its first frontier models, described as state of the art. | Meta | Closed | ||||
| 12 Oct | MiniCPM-V 4.7 MiniCPM-V 4.7 is the latest in OpenBMB's vision-language series, including a 35B-A3B variant that is its first MoE model, plus a 1B version and GGUF quantizations. It handles single-image, multi-image and video understanding. | OpenBMB | Open | 35B-A3B; 1B | |||
| 12 Oct | Muse Wonder Meta releases Muse Wonder and Muse Wonder Pro, two models from its revamped Superintelligence Labs that it claims rival the industry's best, following Muse Spark and its Muse agent. | Meta | Closed | proprietary | |||
| 12 Oct | GLM-Legal A new agentic model from Z.ai built for legal tasks such as document review and drafting, automating multi-step legal workflows. | Z.ai | Closed | ||||
| 12 Oct | Muse Wonder 1.0 A new model called Muse Wonder 1.0 has surfaced on Product Hunt and Reddit with benchmarks, though details on capabilities and availability remain limited. | Muse | |||||
| 12 Oct | Boole-1.7B Boole-1.7B is a 1-bit version of Qwen3-1.7B with 247 MB packed weights, 13.9x smaller than BF16 and scoring 85.9% of the BF16 model across ten tasks. It can be run locally. | Precision Numerics | Open | 1.7B | |||
| 12 Oct | DeepSeek V4.1 Flash DeepSeek V4.1 Flash sees setup guides, spec and price breakdowns, plus MiMo-V2.6-Pro comparisons and AMD MI355X AA-AgentPerf results. | DeepSeek | Open | ||||
| 12 Oct | Claude Opus 5.5 Anthropic's Claude Opus 5.5 ranks first on the Epoch AI leaderboard, an independent benchmark tracker of frontier model performance. | Anthropic | Closed | ||||
| 12 Oct | Falcon-ASR TII releases Falcon-ASR, a 1.6B-parameter Arabic speech recognition model with an Emirati focus, reporting 20.92% average WER on six test sets. Weights and API are not yet available. | TII | 1.6B | ||||
| 12 Oct | ML4 Mistral unveils ML4, an open-weight model that it says gives it an edge over Chinese rivals, as it competes with US labs. | Mistral | Open | ||||
| 12 Oct | Grok 4.7 xAI releases Grok 4.7, with pricing and benchmark comparisons against Anthropic's Claude 5.5 published alongside it. | xAI | Closed | ||||
| 12 Oct | GPT-6.1 Sol OpenAI raises the default context length of GPT-6.1 Sol to 500,000 tokens. | OpenAI | Closed | 500K | |||
| 11 Oct | Decision 3.0 (vllm-sr/d3) Decision 3.0 is a family of five open multimodal models from 0.8B to 27B parameters that answer choice, yes/no and score questions about text and images in one call without generating text. Released under Apache-2.0 on Hugging Face. | vLLM Semantic Router | Open | 0.8B to 27B | Apache-2.0 | ||
| 11 Oct | Claude Haiku 5.5 Claude Haiku 5.5, Anthropic's smaller model, reaches GitHub Copilot, with pricing and best-use guidance published. The source text does not give full pricing or context details. | Anthropic | Closed | 100K | |||
| 11 Oct | OrcaCyber Zero 1.5 OrcaRouter's gated OrcaCyber Zero 1.5 cybersecurity model has a 1M-token context window. It reports 100% on Cybench and 95.8% on an evaluable CVE-Bench subset, priced at $3.00/$7.50 per 1M tokens. | OrcaRouter | Closed | 1M | gated | $3.00 / $7.50 per 1M tokens | |
| 10 Oct | Qwen-Image-2.1 Viggle Turbo v0.3 A distilled version of Qwen-Image-2.1 that generates and edits images in 6 steps without classifier-free guidance, about 5x faster than the 40-step base model. Demo Space available. | Viggle | Open | Qwen Research License (non-commercial) | |||
| 10 Oct | Qwen-Image-2.1 An image model for text-to-image generation and image editing, with native RGBA transparency support. Interactive demo Space is available on Hugging Face, with weights and blog linked. | Qwen | Open | ||||
| 10 Oct | FLUX.2 Klein 9B An experimental web app for editing images with FLUX.2 Klein 9B and multiple LoRAs. Uploaded images are deleted after 7 days, and the author notes uploads may be logged. | Open | 9B | ||||
| 10 Oct | Clef-omni Open Apache 2.0 models that take a state (text, JSON, images or video) and typed questions, returning probabilities for every option in one forward pass. Clef-omni is on Hugging Face. | Cloudflare | Open | 30B-A3B | apache-2.0 | ||
| 10 Oct | Kolibri-1 An open Apache 2.0 mixture-of-experts model focused on German and English, with explicit reasoning mode, tool calling and long-context efficiency. Weights are on Hugging Face. | Aleph Alpha | Open | apache-2.0 | |||
| 10 Oct | EMA Lightning A tiny open Turkish text-to-speech model, about 34 MB in total, with a 5.6M-parameter DiT acoustic model and 3M vocoder. Reports 0.92% WER on Freya-TR-Eval and very fast speech output. | Open | 8.6M | apache-2.0 | |||
| 10 Oct | d1-3B Open decision models that take a state and questions and return typed answers in one forward pass. d1-3B is built on LFM2.5-VL-3B and claims the top score under 10B on the Decision Index. | Liquid AI | Open | 3B | other | ||
| 10 Oct | EmbeddingGemma 2 GGUF Quantized GGUF versions of EmbeddingGemma 2, an open Apache 2.0 embedding model that maps text, code, images, video and audio into one space. Unsloth provides run guides. | Unsloth | Open | apache-2.0 | |||
| 10 Oct | Qwen3.8-27B-Coder390-EfficientThink A community fine-tune of Qwen3.8-27B, post-trained with several rounds of SFT and RLOO and MTP/DFlash variants, available as FP8, NVFP4, INT8 and GGUF weights. | Open | 27B | apache-2.0 | |||
| 10 Oct | Qwen3.8-Flash-Next GSQ-RCO-abliterated A quantized GGUF version of Qwen3.8-Flash-Next where the refusal direction was removed by swapping 144 tensors for abliterated weights. Context is 262K tokens, under Apache 2.0. | Open | 262K | apache-2.0 | |||
| 10 Oct | Qwen-Image-2.1 Outfit Swap Consistency LoRA A LoRA for Qwen-Image-2.1 that dresses a person in a supplied outfit image while keeping the original picture frame, face and background from shifting. | Open | other | ||||
| 10 Oct | Iris-3B A 3-billion-parameter model that generates images directly in pixel space without a VAE. Fine-tuned, it estimates depth and restores and upscales images. Apache 2.0 licensed. | Speri Labs | Open | 3B | apache-2.0 | ||
| 10 Oct | GLM-5.3 UNCENSORED EXL3 3.0bpw A 3.0 bpw EXL3 quantization of a weight-edited GLM-5.3 variant, with 753B total parameters, for running locally. Not affiliated with Z.ai, and MIT licensed. | Open | 753B | mit | |||
| 10 Oct | pplx-decider-v1.1-27b An Apache 2.0 27B decision model built on Qwen3.8-27B that lifts the Decision Index score from 56.4 to 61.56, beating Jev by over 3.5 points using the same backbone. | Perplexity | Open | 27B | apache-2.0 | ||
| 10 Oct | Kimi K3 OpenAI enterprise users can apply existing commitments to Moonshot AI's Kimi K3 through Baseten, according to a Reddit post. Details of the arrangement are limited. | Moonshot AI | Open | ||||
| 10 Oct | MORENA MORENA is a 1.5 billion parameter open-weight language model trained from scratch for 12 African languages. Developers can download the weights and run it themselves. | Vambo AI | Open | 1.5B | |||
| 10 Oct | gpt-rosalind-discovery A gpt-rosalind-discovery model has appeared in OpenAI's API pricing and documentation, with no official announcement yet. Details about its capabilities are not confirmed. | OpenAI | Closed | ||||
| 09 Oct | Microsoft-Decision-1 Microsoft released Microsoft-Decision-1, a decision-scoring model for fast decision-making, built on Alibaba's open-weight Qwen3.5-9B. It is available on Hugging Face and Microsoft Foundry. | Microsoft | Open | 9B |