Model releases

Every A.I. model we have reported on: who built it, how big it is, and whether you can download the weights.

74 models · Updated as we publish · Page 2 of 2

46 with downloadable weights

SearchDeveloper

Weights

74 models

Model	Developer	Size	Weights	Licence	Released	Our coverage
DeepSeek V4-Pro	DeepSeek	—	—	—	August 13, 2026	DeepSeek's Open-Source Agent Harness Passed 233,000 Stars as Its Own API Got PricierSep 22
MiMo-V2.6-Pro	Xiaomi	1.02T total / 42B active MoE	Open	MIT	September 21, 2026	Xiaomi's MiMo-V2.6-Pro Matches Grok 4.7 for $2.62 MillionSep 22 Xiaomi Releases MiMo-V2.6 Weights Under MIT, a 1.02-Trillion-Parameter Pro ModelSep 28
Qwen-Image-2.1	Alibaba	7B (visual generation component)	Open	Qwen Research License Agreement (non-commercial)	September 20, 2026	Qwen-Image-2.1 Ships With Native TransparencySep 21 Alibaba's Qwen-Image-2.1 Drops Apache 2.0 for a Paid Commercial LicenseSep 24 Qwen-Image-2.1 Is Research-Only, Yet an Uncensored Copy Drew 2.19 Million Downloads in a MonthOct 11
NASA-IBM Lunar Foundation Model	NASA and IBM Research	—	Open	Apache-2.0	September 10, 2026	NASA and IBM Open-Source a Moon Model That Warns Against Trusting ItSep 20
CUA-S1-Forms	Cua	706,048	Open	MIT	September 18, 2026	Cua Ships CUA-S1-Forms With a Checkpoint It Can't LoadSep 20
DAMO RADAR	Alibaba DAMO Academy	—	Open	Apache 2.0 (GitHub) / CC BY-NC-SA 4.0 (Hugging Face)	September 18, 2026	Alibaba's DAMO RADAR Beat 23 of 26 Radiologists on CT ScansSep 19
Ternary Bonsai 2 27B	PrismML	27.36B total (24.35B language backbone)	Open	Apache 2.0	September 17, 2026	PrismML Shrinks a 27B Model to 5.9GB at 1.72 BitsSep 18 PrismML's Ternary Bonsai 2 27B Hides a 75% ScoreSep 19 PrismML Shrinks a 27-Billion-Parameter Model to 5.9 GigabytesSep 21 Prism ML Squeezes Qwen3.8-27B Into 5.95 GB at 1.72 BitsSep 29 PrismML's Ternary Bonsai 2 Squeezes Qwen3.8 27B to 5.95 GB, but Stock llama.cpp Won't Run ItOct 1
NeoHorse-1-4B	TokenRhythm	4B (fine-tune of Qwen3.5-4B)	Open	Apache 2.0	September 9, 2026	TokenRhythm Calls a 4B Model a Step Toward Self-Improving AISep 18
Xing4.0-29B-A4B	China Telecom Artificial Intelligence Technology Co.	29B total, 4B active (MoE)	Open	Apache 2.0	September 18, 2026	China Telecom's Xing4.0 Ships After Weeks of Leaked PRsSep 18 China Telecom's Xing4.0 Trains Entirely on Huawei ChipsSep 20 China Telecom's Xing4.0-29B-A4B Fits on a 15GB Desktop GPUSep 24 China Telecom Ships a 29B Model Trained Only on Huawei ChipsSep 27
Edge0-35B-A3B-preview	Edge0-AI	35B MoE, 4 of 256 experts active	Open	Apache 2.0	—	Edge0's 35B Model Runs in Under 3GB of MemorySep 17
WeatherNext 3	Google	—	—	—	September 3, 2026	Google's WeatherNext 3 Forecasts Rain Hourly at 5 KilometersSep 15
Qwen3.8-27B	Alibaba	27B	Open	Apache 2.0	August 17, 2026	Alibaba's Qwen3.8-27B Tops 7.7 Million Hugging Face DownloadsSep 15 Alibaba's Qwen3.8-27B Fits a Frontier Model on One GPUSep 19 Qwen3.8-27B Tops Hugging Face With 7 Million DownloadsSep 29 Qwen3.8-27B Passes 7 Million Downloads Under Apache 2.0, With Training Data UndisclosedSep 30 Qwen3.8-27B Ships Under Apache 2.0 and Fits in 17GB, but Spends 160 Million Tokens Where the Median Spends 43 MillionOct 2 Qwen3.8-27B Has 6.76 Million Downloads Under Apache 2.0, but the 2.4T Flagship Carries a Different LicenceOct 5 Qwen3.8-27B Is Apache 2.0 With a 262,144-Token Window and Already Has 482 Finetunes, Including Cloudflare's ClefOct 7 Qwen3.8-27B Has 6.78M Downloads and an Apache 2.0 Licence, but Its Card Gives Training Data One LineOct 9 Qwen3.8-27B Is Apache 2.0. The 2.4T Max Weights Have No Published Licence TextOct 10 Qwen3.8-27B Reaches 7.1 Million Downloads Under Apache 2.0, on Benchmarks Only Alibaba Has RunOct 14 Qwen3.8-27B Is Apache 2.0, but Artificial Analysis Clocks It at 200M TokensOct 23 Qwen3.8-27B Ships Under Apache 2.0 With 262,144-Token Context, While Its 2.4T Sibling Carries a $50 Million Revenue ClauseOct 26
YuE2-3B	M-A-P	3B	Open	Noncommercial	September 9, 2026	YuE2-3B Beats Suno v5 on Its Own Music BenchmarkSep 14
Gemini 3.8 Flash Cyber	Google	—	Closed	—	—	Gemini's New Cyber Model Found a 13-Year-Old Chrome BugSep 13
Nex-N2.5	Nex-AGI	Up to 1.6T MoE	—	—	—	Nex-AGI's New Agent Model Runs on Qwen's ArchitectureSep 12
Edge0-35b	Edge0	35B MoE	Open	Apache 2.0	—	Edge0 Streams a 35-Billion-Parameter Model Into 3GB of RAMSep 12
LTX-2.5	Lightricks	22B	Open	Free under $10M revenue	August 11, 2026	Lightricks Gates Free Use of LTX-2.5 at $10 Million RevenueSep 12 Lightricks' LTX-2.5 Undercuts Veo on Price, Not on OpennessSep 19 LTX-2.5's Free Commercial Licence Stops at $10 Million in Group Revenue, and Its Card Asks for Contact DetailsOct 8
TimesFM 3.0	Google	330M	Open	Non-Commercial v1.0	—	Google Ships TimesFM-3, Drops the Apache License It Used BeforeSep 12
MiniCPM5-2B	OpenBMB	2.52B	Open	Apache 2.0	September 7, 2026	OpenBMB's 2.5B MiniCPM5 Beats Qwen3.5-4B on Average ScoreSep 11 OpenBMB's 2.5B-Parameter MiniCPM5 Outscores Models Twice Its SizeSep 13
DeepSeek V4.1-Flash	DeepSeek	552B MoE	Open	MIT	September 10, 2026	DeepSeek Ships V4.1-Flash, a 552B MoE Model Under MIT LicenseSep 11 DeepSeek's V4-Pro Survives Past Its Own Retirement DateSep 19 DeepSeek's V4.1 Flash Tops Its Own Charts, Fails MindStudio's TestsSep 20 DeepSeek Ships a 552-Billion-Parameter Model With No VRAM NumbersSep 23
GPT-Live-1	OpenAI	—	Closed	—	September 10, 2026	OpenAI Prices Voice Model GPT-Live-1 at 5 Cents a MinuteSep 10 OpenAI Brings GPT-Live-1 to the API at 5 Cents a MinuteSep 15
SWE-2	Cognition	—	—	—	—	Cognition's SWE-2 Undercuts Fable 5.1 on Price, Not Every BenchmarkSep 10
GPT-6 Astra	OpenAI	—	Closed	—	September 3, 2026	OpenAI's First 'Critical' Model Ships Behind a GateSep 9
Claude Fable 5.1	Anthropic	—	Closed	—	—	Claude Fable 5.1 Has Four Breaking Changes, Not ThreeSep 9

Figures are as stated in each linked article and its sources. A dash means the developer had not disclosed it when we reported.