Open-Weights on Best of AI

AstaBrief 8B

Wed, 04 Nov 2026 00:00:00 +0000

AstaBrief 8B turns a research question and a set of papers into a cited report. Ai2 also added a Fast mode in Asta that uses it.

Beam

Wed, 04 Nov 2026 00:00:00 +0000

Beam is Reflection’s open sparse mixture-of-experts model, with 501B total and 23B active parameters under Apache 2.0. It is built for coding, reasoning, and agentic work.

EmbeddingGemma 2

Wed, 04 Nov 2026 00:00:00 +0000

EmbeddingGemma 2 puts text, code, image, video, and audio into one embedding space. It has 740M parameters and is small enough to run on-device.

Kimi K3.1 Pro

Wed, 04 Nov 2026 00:00:00 +0000

Moonshot AI released Kimi K3.1 Pro on October 23, 2026. It is a 3T-class model with a 1M-token context window, aimed at coding, 3D, financial research, and agentic tasks. A faster Highspeed variant launched the same day.

Kolibri

Wed, 04 Nov 2026 00:00:00 +0000

Kolibri is Aleph Alpha’s open mixture-of-experts model for English and German, with 78B total and 3B active parameters and up to 1M tokens of context. It is aimed at regulated work.

LFM3

Wed, 04 Nov 2026 00:00:00 +0000

LFM3 is Liquid AI’s new model family built on a shared backbone. The first release is six 2B-class models covering text, vision, audio, omni, and decision tasks.

Ling-3.1-flash

Wed, 04 Nov 2026 00:00:00 +0000

Ling-3.1-flash is a hybrid reasoning model from InclusionAI with 560B total and about 25B active parameters and a 1M-token context. It targets agent, search, office, and coding tasks.

Mellum2.1

Wed, 04 Nov 2026 00:00:00 +0000

JetBrains released Mellum2.1 on October 8, 2026, an open 12B mixture-of-experts model for coding agents. It was trained with reinforcement learning in coding environments.

Muse Spark Open

Wed, 04 Nov 2026 00:00:00 +0000

Muse Spark Open is a natively multimodal mixture-of-experts model from Meta, with 816B total and 42B active parameters and a 1M-token context window. It targets coding and agentic work.

Nemotron 3.5 Bolt E4B

Wed, 04 Nov 2026 00:00:00 +0000

Nemotron 3.5 Bolt E4B is a small model that fits in 4GB, built for conversational agents with multi-turn tool use and configurable reasoning. It has a 64K context window.

Nemotron 3.5 Transcribe

Wed, 04 Nov 2026 00:00:00 +0000

Nemotron 3.5 Transcribe follows instructions for transcription, translation, and speaker attribution across 40 languages, and handles recordings up to an hour long.

Qwen3.6 27B

Wed, 04 Nov 2026 00:00:00 +0000

Qwen3.6 27B is a 27B-parameter model from Alibaba’s Qwen team, released October 21, 2026.

WeVisOne-VL

Wed, 04 Nov 2026 00:00:00 +0000

WeVisOne-VL is a vision-language model from Tencent in 2B and 4B sizes. It handles detection, localization, counting, OCR, segmentation, and matting.