Model releases

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

73 models · Updated as we publish · Page 1 of 2

46 with downloadable weights

SearchDeveloper

Weights

73 models

Model	Developer	Size	Weights	Licence	Released	Our coverage
Nanbeige4.3-33B	Nanbeige LLM Lab	33B	Open	Apache-2.0	—	Nanbeige4.3-33B Posts 65.5 on DeepSWE Against Qwen3.8-27B's 42.5, With No Independent Run YetOct 30 Nanbeige4.3-33B Ships Under Apache-2.0 With a 524,288-Token Window and Self-Reported WinsNov 1
LFM3-1.7B	Liquid AI	1.73B	Open	LFM Open License v1.0	October 27, 2026	Liquid AI's LFM3-1.7B Runs at 43.6 Tokens a Second on an iPhone 17 Pro, but Firms Above $10 Million Need a Paid LicenceOct 30
Mistral Large 4	Mistral AI	~1T MoE, 52B active	Open	Mistral Ecosystem Open-Weight License	October 29, 2026	Mistral Large 4 Weights Land With a Clause Barring Hosted Inference Without a LicenceOct 30 Mistral Large 4 Weights Are on Hugging Face, and the Card Disagrees With the Launch PostOct 31
Flint	Recursive Superintelligence	39.8B MoE, about 2.5B active	Open	Apache 2.0	October 21, 2026	Recursive's Flint Posts 89.7% on SWE-bench Pro v2 Under Apache 2.0, With No Outside EvaluatorOct 28
Gemini 4 Carbon	Google	—	—	—	October 28, 2026	Google Opens Gemini 4 Carbon to Cyber Defenders Only and Claims 59 on Artificial AnalysisOct 28
Hy-Nomas	Tencent	—	Open	Apache-2.0	October 21, 2026	Tencent's Hy-Nomas World Model Is Apache-2.0 With 196 GB of Weights and No Published ScoresOct 27
Qwen3.8-Flash-Next	Alibaba	125B main, 6B active, plus 51B N-gram embeddings	Open	Qwen Community License 1.0	August 26, 2026	Cohere Says a Qwen Model Tried to Upload a Test String to an Alibaba Bucket From Its SandboxOct 27
MiniMax-M3.1-Flash	MiniMax	428B MoE, 23B active (model card)	Open	MiniMax Community License	October 19, 2026	MiniMax-M3.1-Flash Has a Downloadable 428B Repository, While Artificial Analysis Calls It Proprietary With Weights "Not Publicly Available"Oct 26 MiniMax's M3.1-Flash Weights Come With a $20 Million Revenue Line and a Military-Use BanOct 29
GPT-6.1 Terra	OpenAI	—	—	—	October 26, 2026	OpenAI's GPT-6.1 Terra Costs $1 per Million Input Tokens and Scores 50.3% on Terminal-Bench 4.0, Behind Sonnet 5.5's 70.6%Oct 26
Kimi K3	Moonshot AI	2.8T MoE, 104B active	Open	Custom Moonshot licence (revenue-tiered)	July 27, 2026	Cerebras Says Kimi K3 Fits on 59 Wafers, and Gives No Price for Any of ThemOct 25
Dream-1	Dream Labs	—	—	—	October 19, 2026	Dream Labs Sells Robot Control at $3 an Hour After $90M SeedOct 24
Kimi K3.1 Pro	Moonshot AI	2.8T MoE, 104B active	—	—	October 24, 2026	Moonshot Filters Cyber Data From Kimi K3.1 Pro and Will Release Open Weights by Nov. 4Oct 24
WorldTUI	Simon Edwardsson	360M	Closed	—	October 21, 2026	WorldTUI Is a 360M-Parameter Model That Fakes a Linux Terminal, and Its Weights Are Not Out YetOct 22
Qwen3-4B-deslop v4	Erik Markus Kannike	4B	Open	Apache-2.0	—	Qwen3-4B-Deslop Cuts Its Slop Score From 121 to 12.5 on a MacBook Air, and Two of 40 Answers Came Out DamagedOct 19
HybridAgent-14B-Agentic	Together AI	14B	Open	MIT	October 13, 2026	HybridAgent Paper Claims 10.4% Gain and 2.7x to 3.7x Lower Cost, but Its Own Text Gives 10.4% Two Different MeaningsOct 19
Glimpse	Elorian	—	—	—	October 16, 2026	Elorian's Glimpse Costs 94% Less Than GPT-6 Astra and Trails the Best Rival on 7 of 12 TestsOct 17 Elorian Launches Glimpse, a Visual Model Priced at $0.60 per Million Tokens, Scored Only by ElorianOct 20
DeepSeek-V4.1-Flash-1017	DeepSeek	763B listed in Safetensors metadata	Open	MIT	October 17, 2026	DeepSeek-V4.1-Flash-1017 Lifts Terminal-Bench 4.0 From 31.2 to 42.3 Under the Same MIT LicenceOct 17 DeepSeek's Flash-1017 Gains 11.1 Points on Terminal-Bench 4.0 in Its Own Table, and the Card Gives No Release DateOct 19 DeepSeek Swapped V4.1-Flash-1017 Behind Its API Name, and the New Card Lists NL2Repo at 65.4 Where the Old One Said 64.0Oct 19 DeepSeek Claims 42.3 on Terminal-Bench 4.0 for V4.1-Flash-1017; Vals AI Measured 31.8Oct 21 DeepSeek V4.1-Flash-1017 Posts Gains on All 11 Listed Benchmarks, Every One Of Them Its OwnOct 25 DeepSeek-V4.1-Flash-1017 Is MIT-Licensed at 763B and Beats Its Predecessor on All 11 Benchmarks, by DeepSeek's CountOct 27
Nemotron 3.5 Bolt 74B-A9B	Nvidia	74B MoE, 9B active	Open	OpenMDW-1.1	October 15, 2026	Nvidia Releases Nemotron 3.5 Bolt, a 74B Vision Model Packed Into 24GB at 2 BitsOct 16
MAI-Base-2	Microsoft AI	2.8T MoE, 80B active	Closed	—	—	Microsoft AI Says MAI-Base-2 Has 2.8T Parameters and Trained on 16,000 GB200s in 38 DaysOct 16
Krea 3	Krea	—	—	—	October 15, 2026	Krea 3 Goes Live on fal With 8-Image Editing at $0.025 Per MegapixelOct 16
Muse Spark Open	Meta	816B total, 42B active	Open	Muse Model License	October 15, 2026	Meta Releases Muse Spark Open, 816B Parameters Under a Licence It Calls LimitedOct 15 Apple Runs Meta's 816B Muse Spark Open on Four Mac Studios at 36.7 Tokens a Second, and Its Own Page Gives Two First-Token TimesOct 17 Meta's Muse Spark Open Weights Come With a $20 Million Resale Threshold and a Required Credit LineOct 18 Meta's Muse Spark Open Weights Come With a $20 Million Revenue Clause and a 700 GB Memory FloorOct 20 Meta Releases Muse Spark Open, an 816B Model, With a $20 Million Revenue Trigger in the LicenseOct 22
LFM2.5-24B-A2B	Liquid AI	24B MoE (2B active)	Open	LFM Open License v1.0	October 13, 2026	Liquid AI Releases LFM2.5-24B-A2B With a $10 Million Revenue Cap on Free Commercial UseOct 15 Liquid AI's LFM2.5-24B-A2B Beats Qwen3.6-35B-A3B on Agent Tests and Trails It on CodingOct 17 Liquid's LFM2.5-24B-A2B Licence Ends at $10 Million in Revenue, and Its Table Scores Qwen Below Qwen's Own CardOct 19 Liquid AI's LFM2.5-24B-A2B Scores 27.69 on AutomationBench, and Its Post Admits the Model Cheated in TrainingOct 20
MiniCPM-V 4.7 35B-A3B	OpenBMB	35.2B MoE	Open	Apache-2.0	October 6, 2026	OpenBMB's MiniCPM-V 4.7 Went Up With No Model Card on Oct. 6, and Now Reads Apache-2.0 With No Benchmark NumbersOct 14 MiniCPM-V 4.7 Went Up Oct. 6 With No Model Card; Its 3 GB iPhone Claim Still Lacks a CheckpointOct 17 MiniCPM-V 4.7 Is Apache 2.0 With 3B Active Parameters and No Benchmark TablesOct 18 MiniCPM-V 4.7 Ships 35B Weights Under Apache-2.0, With Its Benchmark Scores Left in ChartsOct 20
Muse Wonder	Meta	—	Closed	Proprietary	October 12, 2026	Meta's Muse Wonder Ships Closed at $4 and $20 Per Million Tokens, Ranked Ninth by Artificial AnalysisOct 14
d1-3B	Liquid AI	3.12B	Open	LFM 1.0	October 7, 2026	Liquid AI Opens d1-3B, a Decision Model That Answers in 8 ms, With Its Own ScoreboardOct 11
Microsoft-Decision-1	Microsoft	Post-trained Qwen3.5-9B	—	—	October 9, 2026	Microsoft-Decision-1 Is a Post-Trained Qwen3.5-9B at $0.042 per Million Tokens, With No Licence GivenOct 10 Microsoft's Decision-1 Is a Post-Trained Qwen3.5-9B at $0.042 Per Million Tokens, With Self-Reported BenchmarksOct 26
Claude Haiku 5.5	Anthropic	—	Closed	Proprietary	October 7, 2026	Claude Haiku 5.5 Costs $0.10 Per Million Input Tokens, a Tenth of Haiku 4.5, and Scores 39.2% on Terminal-Bench 4.0Oct 7 Claude Haiku 5.5 Launches at $0.10 per Million Input Tokens, the Same List Price as GPT-6 LunaOct 8 Claude Haiku 5.5 Costs $0.10 Per Million Input Tokens, Then Writes 440 Million Tokens to Finish One BenchmarkOct 14
EmbeddingGemma 2	Google DeepMind	740M	Open	Apache 2.0	October 6, 2026	Google's EmbeddingGemma 2 Packs Text, Image, Video and Audio Into 740M Parameters Under Apache 2.0Oct 7 EmbeddingGemma 2 Embeds Text, Images, Audio and Video in 567MB of RAM on a Pixel 11 ProOct 8 Google's EmbeddingGemma 2 Adds Images, Video and Audio to a 740M EmbedderOct 9 Google's EmbeddingGemma 2 Embeds Text, Images, Audio and Video in 740M Parameters, but English Retrieval SlippedOct 11 EmbeddingGemma 2 Maps Text, Video and Audio Into One 768-Dimension SpaceOct 13 EmbeddingGemma 2 Covers Five Modalities in 740 Million Parameters, and 128-Dimension Vectors Cost It 13 PointsOct 15 EmbeddingGemma 2 Lifts MTEB Code to 78.68 From 68.76, but Multilingual Text Moves 0.21 PointsOct 19
Beam	Reflection AI	501B MoE, 23B active	Open	Apache 2.0 (announced, weights not yet released)	October 5, 2026	Reflection's Beam Has 501B Parameters and 23B Active, and Its Weights Are Still Due "Later This Month"Oct 5 Reflection Announces Beam, a 501B Open-Weight Model, but Has Not Named a LicenceOct 6 Reflection's 501B Beam Model Gets Apache 2.0 Weights on Hugging FaceOct 16
Holo4-27B	H Company	27B dense	Open	CC BY-NC 4.0	September 28, 2026	H Company's Holo4-27B Scores 61.7% on OSWorld 2.0, and Its Model Card Bars Commercial Use of an Apache 2.0 BaseOct 5
Kolibri	Aleph Alpha	78.1B MoE, 3.46B active	Open	Apache 2.0	October 3, 2026	Aleph Alpha's Kolibri Ships Under Apache 2.0, Compared Only With Spring ModelsOct 4
Praxis-1	Runway	—	—	—	—	Runway's Praxis-1 Robot Model Is Open-Weight on Paper, With Weights Still UnreleasedOct 4 Runway's Praxis-1 Robot Model Claims 16.1 cm Placement Error From Web Video, and Its Weights Are Not Out YetOct 5
Kolibri 1	Aleph Alpha	78.1B MoE, 3.46B active	Open	Apache 2.0	October 3, 2026	Aleph Alpha Releases Kolibri, a 78B German-English Model, Under Apache 2.0Oct 3 Aleph Alpha's Kolibri-1 Is a 78B Apache 2.0 Model That Activates 3.46B Parameters, and Its Training Data Stays ClosedOct 6 Kolibri-1 Is Apache 2.0 and Fits on One B200, but Trails Qwen3.8 27B by 9.1 Points in GermanOct 8
CLM-8B	Contrastive-LM	8B	Open	Apache-2.0	—	Contrastive-LM Releases CLM-8B, an Apache 2.0 Verifier Claiming 87.6% on Terminal-BenchOct 3
Clef	Cloudflare	27B	Open	Apache-2.0	October 1, 2026	Cloudflare Releases Clef, a 27B Decision Model, Under Apache 2.0Oct 3 Cloudflare Releases Clef, a 27B Model That Returns Probabilities Instead of TextOct 4 Cloudflare's Clef Reports 209 ms Decisions, but the Outside Tests of Decision Models So Far Cover Only JevOct 6 Cloudflare's Clef Is a 27B Apache 2.0 Model That Never Writes Text, and Its Own Table Has the Baseline Winning Two of Six RowsOct 8 Cloudflare's Clef Beats TypeSafe's Jev on Three Tests and Loses on OneOct 9 Cloudflare Open-Sources Clef, a 27B Drop-In for TypeSafe's JevOct 10
Strands Decider 2B	Amazon Web Services	about 2B	Open	Apache 2.0	October 1, 2026	Amazon Releases Strands Decider 2B, an Apache 2.0 Decision Model Built on Qwen3.5-2BOct 1
Gemini 4 Argon	Google DeepMind	—	Closed	Proprietary	September 30, 2026	Google's Gemini 4 Argon Goes to Cyber Defenders First, at $2 Per Million Input TokensOct 1 Gemini 4 Argon Leads 13 of 18 Benchmarks Google Chose, but Only Cyber Defenders Can Use ItOct 2 Google Gives Gemini 4 Argon to Cyber Defenders First, Sets No Public DateOct 3 Gemini 4 Argon Goes to Cyber Defenders First, With Broad Access UndatedOct 4 Gemini 4 Argon Costs $1.99 a Task at Promo Price Against $0.72 for GPT-6.1 Sol, and Is Not Yet on SaleOct 8 Gemini 4 Argon Goes Wide at $2 and $10 Per Million Tokens Until Nov. 30, Then DoublesOct 13 Gemini 4 Argon Reaches Paying Developers at $2 Per Million Tokens, Trailing Rivals on Four BenchmarksOct 16 Gemini 4 Argon Opens to the API at $2 per Million Tokens, and Loses 5 of 19 Rows in Google's Own TableOct 18 Gemini 4 Argon Reaches the API at $2 and $10 Per Million Tokens, With a No. 1 Ranking and No ScoreOct 19 Gemini 4 Argon Reached General Availability October 13 at a $2 Promotional Price, After a Cyber-Partners-First LaunchOct 26 Gemini 4 Argon Ranks Third on Vals, Not First as Google's Card Says, and Its API Price Doubles on Dec. 1Oct 27
ZDTaichu5.0-9B	TaichuAI	9B	Open	NVIDIA Open Model License Agreement	—	ZDTaichu5.0-9B Ships Under an NVIDIA Licence Its Hugging Face Tags Omit, and Its Agent Scores Mix Test SetupsSep 30
CLM-v0.1-8B	Contrastive-LM	Frozen Qwen3-8B encoder plus about 20M trainable head parameters	Open	Apache 2.0	September 21, 2026	CLM-v0.1-8B Puts an Apache 2.0 Rival to TypeSafe's Jev on Hugging Face, With 30-Task Test SetsSep 30
GPT-6.1 Sol	OpenAI	—	Closed	—	September 29, 2026	GPT-6.1 Sol Is One-Fifth of Astra's Price, but Only the Cached Rate Is NewSep 29 GPT-6.1 Sol Costs One-Fifth of Astra at $2 and $10 per Million Tokens, on OpenAI's Own TestsSep 30 GPT-6.1 Sol Lands One Point Behind GPT-6 Astra on Artificial Analysis IndexOct 3 GPT-6.1 Sol Is Priced at One-Fifth of Astra, and Its System Card Rates Cyber CriticalOct 4 GPT-6.1 Sol Costs $2 Input and $10 Output Per Million Tokens, and OpenAI Rates It Critical in CybersecurityOct 6
Nemotron 3 Diarization	Nvidia	100M	Open	OpenMDW License 1.1	September 23, 2026	Nvidia's Nemotron 3 Cuts Speaker-ID Errors by 41%Sep 28 Nvidia's 100M-Parameter Nemotron 3 Diarization Tops VoiceArena at 14.72% ErrorOct 3
Hemmingway-1	Altworld	27B	Open	CC BY-NC 4.0 (noncommercial)	September 21, 2026	Altworld's Hemmingway-1 Isn't Apache-Licensed, Despite ReportsSep 28
Bonsai 2 27B	PrismML	27B, ternary weights at 1.76 effective bits each	Open	Apache 2.0	September 17, 2026	PrismML Shrinks Qwen3.8 27B to 5.9GB, Keeps 98% of Its ScoreSep 27
Audio8 ASR Infinite	Edge0-AI	4B	Open	—	September 21, 2026	Edge0's 'Open' Speech Model Has a Non-Commercial Decoder InsideSep 27
Mercury 2.5	Inception	—	Closed	proprietary	September 8, 2026	Inception's Mercury 2.5 Hits 770 Tokens a SecondSep 24
Gemini 3.8 Flash TTS	Google	—	Closed	proprietary	September 23, 2026	Gemini 3.8 TTS Tops a Benchmark Its Own Hire BuiltSep 23 Google's Gemini 3.8 TTS Pricing Doubles on Jan. 1, 2027Sep 25
GPT-6 Sol	OpenAI	—	Closed	—	September 22, 2026	OpenAI Cuts GPT-6 Sol and Luna Prices in HalfSep 23
Claude Opus 5.5	Anthropic	—	Closed	—	September 22, 2026	Anthropic's Claude Opus 5.5 Matches Fable 5.1 at 40% Lower CostSep 23
Jev	TypeSafe AI	—	Closed	—	September 15, 2026	TypeSafe Raises $40 Million for Jev, a Model That Won't TalkSep 22 TypeSafe's Jev Skips Text and Returns a Single Floating-Point NumberSep 23
DeepSeek V4-Pro	DeepSeek	—	—	—	August 13, 2026	DeepSeek's Open-Source Agent Harness Passed 233,000 Stars as Its Own API Got PricierSep 22

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