52 open source AI models from 25 labs. Input prices run from $0.03 to $1.40 per 1M tokens. The largest context window is 1M tokens (DeepSeek-V4 Flash). Newest check: 12 Nov 2026.
94 of 197 models were checked against the vendor's own page in the last 30 days.
| Model | Lab | Context | In / out per 1M | Reads and writes | Status | Checked |
|---|---|---|---|---|---|---|
| Ripple-65B PreviewRicursive Intelligence's 65B dense open-weight preview for agentic coding, with a 1M-token context window. | Ricursive Intelligence | 1M | $0.10 / $0.30 | text → text, code | Preview | 11 Nov |
| DeepSeek-V4.1-Flash-1017DeepSeek's October 2026 update to V4.1-Flash: MIT-licensed weights, image input and a 1M-token context, tuned for coding agents and long task loops. | DeepSeek | 1M | $0.15 / $0.60 | text, image → text | Available | 20 Oct |
| BeamReflection AI's open-weight sparse MoE for coding and agents, with a 1M-token context window and 128K output, under a permissive Apache licence. | Reflection AI | 1M | $1 / $3 | text → text | Available | 19 Oct |
| Nemotron Labs Dory 1.0 4BNVIDIA's open 4B looped model: one checkpoint reuses its shared layers 1, 2 or 4 times per token, with a 256K-token context window and a permissive licence. | NVIDIA | 262K | -- | text, code → text | Available | 20 Oct |
| Hy-Memory-30B-A3BTencent's open-weights memory-writer MoE with 30B parameters, 3B active and a 262K window, released October 2026 for Mem0, MemOS and Hy Memory. | Tencent | 262K | -- | text → text | Available | 2 Nov |
| Nemotron 3.5 Bolt 74B-A9B 24GBNVIDIA's open 74B-A9B vision-language MoE, quantized to 2 bits to fit 24GB, built for local agents. Released in October 2026 with 256K context. | NVIDIA | 262K | -- | text, image, video → text | Available | 20 Oct |
| Nemotron 3.5 Super VL 120B-A12BNVIDIA's open mixture-of-experts vision-language model for document, screenshot and video agents, released in October 2026. | NVIDIA | 1M | $0.10 / $0.40 | text, image, video → text | Available | 20 Oct |
| Mellum2.1Mellum2.1 is JetBrains' open mixture-of-experts model, post-trained with reinforcement learning to work inside code repositories as an agent. | JetBrains | 131K | -- | text → text | Available | 20 Oct |
| EmbeddingGemma 2Google DeepMind's open 740M-parameter embedding model maps text, code, images, video, and audio into one shared embedding space for on-device search. | 8K | -- | text, code, image, audio, video → embeddings | Available | 28 Oct | |
| KolibriKolibri is Aleph Alpha's Apache 2.0 English-German mixture-of-experts model: 78.1B parameters, 3.46B active, released 3 October 2026. | Aleph Alpha | 1M | -- | text → text | Available | 5 Oct |
| Cloudflare ClefCloudflare Clef (27B) and Clef-flash (9B): open-weight decision models from 2026 that return probability-scored answers from text, image and video input. | Cloudflare | 66K | $0.24 / $0 | text, image, video → text | Available | 16 Oct |
| Strands Decider 2BAmazon Strands Labs open decision model (Oct 2026): Apache 2.0, Qwen3.5-2B base, 72.3% on JevBench v19 (vendor-reported). | Amazon | -- | -- | text, image → text | Available | 16 Oct |
| Index-Translate-35B-A3B-previewOpen MoE translation model from Bilibili with 35B total and 3B active parameters, image translation and downloadable weights, released in 2026. | Bilibili | 262K | -- | text, image → text | Preview | 2 Nov |
| MiMo-V2.6-ProXiaomi's MiMo-V2.6-Pro is a 1.02T-parameter open-weights MoE model (Sept 2026) with 1M-token context, MIT license, and $0.435 per 1M input tokens. | Xiaomi Corporation | 1M | $0.435 / $0.87 | text, image, video, audio → text | Available | 22 Sep |
| MiMo-V2.6-Distill-Qwen-9BA 9B dense model, supervised fine-tuned on MiMo-generated agentic data, released by Xiaomi in September 2026 under the MIT license, self-host only. | Xiaomi Corporation | -- | -- | text, image → text | Available | 28 Sep |
| MiMo-V2.6-FlashXiaomi's open-weight MoE model (309B total, 15B active parameters), released in September 2026 with a 1M-token context window and an MIT license. | Xiaomi Corporation | 1M | $0.14 / $0.28 | text, image, video, audio → text | Available | 22 Sep |
| MiMo-V2.6-Flash-RLXiaomi's open-weight 309B-parameter (15B active) MoE model with 1M-token context, priced far below its Pro-RL sibling, released Sept 2026 under MIT. | Xiaomi Corporation | 1M | $0.14 / $0.28 | text, image → text | Available | 28 Sep |
| MiMo-V2.6-Pro-RLXiaomi's open-weight 1.02T-parameter (42B active) MoE reasoning model with 1M-token context, released September 2026 under the MIT license. | Xiaomi Corporation | 1M | $0.435 / $0.87 | text, image, video, audio → text | Available | 28 Sep |
| Atria Dawn PreviewAtria Dawn Preview is Shanghai AI Lab's 744-billion-parameter agentic model, released as an open, MIT-licensed weight download in September 2026. | Shanghai AI Lab | 256K | -- | text → text | Preview | 9 Oct |
| DeepSeek-V4.1-FlashThe MIT-licensed vision-and-text model, launched in September 2026 and refreshed with an October snapshot. | DeepSeek | 1M | $0.15 / $0.60 | text, image → text | Available | 19 Oct |
| Hy4 PreviewHy4 Preview is a 770B-parameter open-weight MoE model from Tencent, released August 2026 with a flagship-class context window and Apache-licensed weights. | Tencent | 1M | $0.834 / $2.50 | text → text | Preview | 2 Sep |
| Ling 3.0 Flash FinA 124B-parameter open MoE model from InclusionAI, continued-trained on financial data for source-grounded investment research and spreadsheet work. | InclusionAI | 262K | Free | text → text | Available | 9 Sep |
| GLM-5.3-FlashOpen-weight, natively multimodal MoE model from Z.ai with a 1M-token context window and MIT license, released in August 2026. | Z.ai | 1M | $0.15 / $0.50 | text, image, video, pdf → text | Available | 29 Sep |
| Qwen3.8-27BA dense 27B vision-language model from Alibaba's Qwen team, open-weight since 2026 with native text, image, and video understanding. | Alibaba Cloud | 262K | $0.50 / $3 | text, image, video → text | Available | 23 Sep |
| DeepSeek-V4-Pro-0813DeepSeek-V4-Pro-0813 is DeepSeek's flagship 1.6T-parameter mixture-of-experts model, generally available since August 2026 with MIT-licensed open weights. | DeepSeek | 1M | $0.66 / $1.98 | text → text | Available | 9 Nov |
| North Micro Vision InstructNorth Micro Vision Instruct is Cohere Labs' 2.4B open-weight vision-language model, built for native-resolution document, chart, and OCR understanding. | Cohere | 128K | Free | text, image → text | Available | 7 Oct |
| Muse GlimmerMuse Glimmer, released August 10, 2026 by Meta AI, is a 30B open-weight agentic model built to run on a single consumer GPU. | Meta AI | 131K | -- | text, image → text | Available | 28 Oct |
| Wan-Animate-2Wan-Animate-2 is Alibaba's open-weight 14B model for character animation, released in 2026, feeding driving video straight into a diffusion transformer. | Alibaba Cloud | -- | -- | image, video, text → video | Available | 28 Sep |
| Shieldstral 1.0Mistral's open-weight multimodal safety classifier, released 4 August 2026, that screens text and images against policies you write. | Mistral AI | 32K | -- | text, image → text | Preview | 12 Nov |
| K-EXAONE 2.0Open-weight mixture-of-experts model from LG AI Research, released in July 2026, with ten supported languages and a hybrid attention design. | LG AI Research | 262K | -- | text → text | Available | 12 Oct |
| Hy3Superseded by Hy4 on 4 November 2026. Hy3 is Tencent's 295B open-weight MoE with a 256K context window; its API is listed at 1 RMB input and 4 RMB output per 1M tokens. | Tencent | 256K | CNY 1 / CNY 4 | text → text | Available | 9 Nov |
| Pulsar 16BPulsar 16B compresses NVIDIA Nemotron 3 Nano 30B to 16.15B parameters, scoring 87.22 on AIME 2025 and 71.41 on GPQA Diamond. Apache 2.0, free to self-host. | Multiverse Computing | 1M | -- | text → text | Available | 30 Aug |
| GLM-5.2GLM-5.2 is Z.ai's June 2026 open-source 744B MoE model with a 1M-token context window, 85.6% GPQA Diamond (Vals AI), and MIT license at $1.40/1M input tokens. | Z.ai | 1M | $1.40 / $4.40 | text → text | Available | 7 Oct |
| BosunBosun v1.1 by Hanno Labs is an Apache 2.0 judge model that scores whether pairs of findings satisfy a runtime-defined rule, available in 4B and 0.6B sizes. | Hanno Labs | 8K | -- | text → text | Available | 31 Aug |
| DiffusionGemma 26B-A4BDiffusionGemma 26B-A4B by Google DeepMind: open-source text diffusion model, 1,100 tok/s on H100, 256K context, 73.2% GPQA Diamond. Free under Apache 2.0. | 256K | -- | text, image, video → text | Available | 27 Sep | |
| Nemotron 3.5 ASRNVIDIA's open streaming speech recognizer, released 4 June 2026, transcribing dozens of language-locales from a single checkpoint. | NVIDIA | -- | -- | audio → text | Available | 2 Nov |
| Gemma 4 12BGemma 4 12B is Google's June 2026 open-weight multimodal model with a 256K context window and native audio/video input on 8GB of VRAM. | 262K | -- | text, image, audio, video → text | Available | 1 Sep | |
| Cohere Command A+Cohere Command A+ (May 2026) is a 218B-parameter open-weights MoE model with 128K context, 48-language support, and Apache 2.0 licensing. | Cohere | 128K | -- | text, image → text | Available | 30 Sep |
| DeepSeek V4DeepSeek V4-Pro reached GA in August 2026 under MIT, matching frontier models on agentic coding. Open-source MoE, 1M-token context, peak/off-peak API pricing. | DeepSeek | 1M | $0.66 / $1.98 | text, image, audio, video → text | Available | 29 Oct |
| DeepSeek-V4-ProDeepSeek V4 Pro: an MIT-licensed frontier-scale MoE, generally available since August 2026, with 1M-token context and text-only input. | DeepSeek | 1M | $1.32 / $3.96 | text → text | Available | 23 Oct |
| Qwen3.6-27BQwen3.6-27B, released April 22, 2026 by Alibaba's Qwen Team, is a 27B open-weight dense model with a 262,144-token context window. | Alibaba Cloud | 262K | -- | text, image, video → text | Available | 28 Oct |
| Gemma 4 31BGemma 4 31B, released April 2, 2026 by Google DeepMind, is a 31B open-weight multimodal model with a 262,144-token context window. | 262K | -- | text, image → text | Available | 16 Sep | |
| ALIA-40b-instruct-2601BSC's open instruct model for Spanish, Catalan, Basque and Galician, post-trained with supervised and preference data and released as Apache-licensed weights. | Barcelona Supercomputing Center | 164K | -- | text → text | Beta | 12 Oct |
| Olmo 3.1 32B ThinkDense 32.2B reasoning model from Ai2 that ships open weights alongside its training data and code, with a 65,536-token window. | Ai2 | 66K | -- | text → text | Available | 3 Oct |
| Ministral 3 8BMinistral 3 8B by Mistral AI (Dec 2025): Apache 2.0, 8B params, 256K context, 70.6% MMLU-Pro, 78.7% AIME 2025 (reasoning), vision input, $0.15/1M tokens. | Mistral AI | 256K | $0.15 / $0.15 | text, image → text | Available | 30 Aug |
| Ministral 3 14BMinistral 3 14B is Mistral AI's flagship small open-weight model, with a dedicated reasoning variant built for math and science benchmarks. | Mistral AI | 256K | $0.20 / $0.20 | text, image → text | Available | 30 Oct |
| Ministral 3 3BMinistral 3 3B by Mistral AI (Dec 2025): Apache open-weight edge model, 3.4B params, vision input, fits in 8GB VRAM. | Mistral AI | 256K | $0.10 / $0.10 | text, image → text | Available | 30 Aug |
| Mistral Large 3Mistral Large 3 (Dec 2025) is a 675B MoE model (41B active) with 256K context and Apache 2.0 open weights. | Mistral AI | 256K | $0.50 / $1.50 | text, image → text | Available | 26 Sep |
| Phi-4Phi-4 is Microsoft's 14B dense model under the MIT license, scoring 84.8% on MMLU in the vendor's own table. It runs locally through Ollama or as a serverless Azure Foundry endpoint. | Microsoft | 16K | $0.125 / $0.50 | text → text | Available | 29 Sep |
| Nemotron 3.5 LightningNVIDIA's open MoE model for AI agents: 3B of 30B parameters active per token, a 1M-token window and quicker agent tasks than a same-size Qwen rival on PinchBench. | NVIDIA | 1M | $0.025 / $0.12 | text, code → text | Available | 6 Nov |
A person checks every record against the vendor's own page. Rankings cannot be bought. How Hokai checks records. Not sure what you need? Describe the job to Smart Match and get three picks with reasons.