Early build · scores computed by the deterministic engine from real, sourced developments (30 days) · weights v1.0-prior
Live
☰
Latest › Models › Compare › Open source
Vikshy Compare

Open-weight (open-source) AI models

AI models released as open weights, the ones you can download and run yourself rather than only calling through an API. 61 in the catalogue, strongest tier first. "Open weights" means the trained model is downloadable; licences vary, so check each lab's terms.

ModelProviderContextInput /MOutput /MAccess
Hunyuan Hy3FrontierCost-efficient frontier reasoning and agent work, positioned against other top open models.
TTencent (Hunyuan)
256K
$0.14
$0.58
API (model id: hy3) via Tencent Cloud TokenHub, plus open weights on Hugging Face and GitHub. Also in Yuanbao, CodeBuddy, and ima.
MiniMax-M3FrontierFrontier coding, agentic execution, and multimodal reasoning at low cost.
MMiniMax
1M
$0.30
$1.20
API (model id: MiniMax-M3) via platform.minimax.io, plus open weights on Hugging Face.
GLM-5.2FrontierFrontier agentic coding and long-context reasoning at a fraction of closed-model cost.
ZZhipu AI (GLM)
1M
$1.40
$4.40
API (model id: glm-5.2) via Z.ai, plus the GLM Coding Plan and open weights on Hugging Face.
Kimi K3FrontierComplex multi-step agents, deep research, and large-codebase work where the top tier matters.
KMoonshot AI (Kimi)
1M
$3.00
$15.00
API (model id: kimi-k3) at api.moonshot.ai, plus kimi.com and the Kimi apps. Open weights planned on Hugging Face.
Command A+FrontierEnterprises that need top quality with self hosting options.
CCohere
128K
see pricing
see pricing
API + open weights
Mistral Large 3FlagshipTeams wanting a top open weight generalist with a hosted option.
Mistral AI
256K
$0.50
$1.50
API + open weights
Kimi K2.6FlagshipTeams wanting a capable open-weight agentic model with generous context at a moderate price.
KMoonshot AI (Kimi)
256K
$0.95
$4.00
API (model id: kimi-k2.6) plus open weights on Hugging Face (moonshotai/Kimi-K2.6).
DeepSeek-V4-ProFlagshipHard reasoning, long documents and agentic tasks.
DeepSeek
1M
$1.74
$3.48
API + open weights
Command AFlagshipEnterprise RAG, agents and general chat.
CCohere
256K
$2.50
$10.00
API + open weights
Llama 4 MaverickFlagshipGeneral assistant and multimodal workloads run through third-party hosts or self-hosting.
Meta
1M
Open
Open
Open weights + third-party API
MiniMax-M1ReasoningLong-context reasoning and software tasks that need a big open thinking model.
MMiniMax
1M
$0.40
$2.20
API (model id: MiniMax-M1) via platform.minimax.io, plus open weights on GitHub and Hugging Face (Apache 2.0).
Magistral SmallReasoningLocal reasoning and cost controlled deployments.
Mistral AI
128K
Open
Open
API + open weights
Kimi K2 ThinkingReasoningMath, planning, and step-by-step problems that benefit from deliberate reasoning.
KMoonshot AI (Kimi)
256K
see pricing
see pricing
API (model id: kimi-k2-thinking) plus open weights on Hugging Face (moonshotai/Kimi-K2-Thinking).
GLM-5.1ReasoningReasoning and coding work on a recent GLM-5 checkpoint.
ZZhipu AI (GLM)
1M
see pricing
see pricing
API (model id: glm-5.1) via Z.ai, plus open weights on Hugging Face.
DeepSeek-V4-FlashBalancedEveryday chat, coding and high volume production use.
DeepSeek
1M
$0.14
$0.28
API + open weights
GLM-4.6VBalancedMultimodal reasoning and document or UI understanding with tool use.
ZZhipu AI (GLM)
128K
$0.30
$0.90
API (model id: glm-4.6v) via Z.ai, plus open weights on Hugging Face.
Kimi K2.5BalancedEveryday agentic and coding work where you want good value on a proven open model.
KMoonshot AI (Kimi)
256K
$0.60
$3.00
API (model id: kimi-k2.5) plus open weights on Hugging Face (moonshotai/Kimi-K2.5).
Mistral Small 4BalancedHigh throughput budget tasks that still need long context.
Mistral AI
256K
see pricing
see pricing
API + open weights
kimi-latestBalancedApps that want the newest Kimi without pinning a version, including vision inputs.
KMoonshot AI (Kimi)
128K
see pricing
see pricing
API only (model id: kimi-latest). Hosted, no open weights for the alias.
GLM-4.6V-FlashEfficientOn-device or high-volume multimodal tasks at no token cost.
ZZhipu AI (GLM)
128K
Free
Free
API (model id: glm-4.6v-flash) via Z.ai, free tier, plus open weights on Hugging Face.
Command R7BEfficientHigh volume lightweight generation.
CCohere
128K
$0.0375
$0.15
API + open weights
Llama 4 ScoutEfficientVery long context tasks and efficient on device or single GPU deployment.
Meta
10M
Open
Open
Open weights + third-party API
Ministral 3 14BEfficientOn device and latency sensitive text tasks.
Mistral AI
128K
Open
Open
API + open weights
Voxtral MiniEfficientVoice interfaces and transcription pipelines.
Mistral AI
32K
Open
Open
API + open weights
GLM-4.5-AirEfficientBudget-sensitive apps that still want GLM-4.5-class output.
ZZhipu AI (GLM)
131K
see pricing
see pricing
API (model id: glm-4.5-air) via Z.ai, plus open weights on Hugging Face.
Devstral SmallOpenSelf hosted coding agents on a budget.
Mistral AI
256K
$0.10
$0.30
API + open weights
Pixtral 12BOpenImage plus text apps that want open weights.
Mistral AI
128K
$0.15
$0.15
API + open weights
MiniMax-M2.1OpenSelf-hosting or API use where you want an updated open M2-class model.
MMiniMax
204K
$0.30
$1.20
API (model id: MiniMax-M2.1) via platform.minimax.io, plus open weights on Hugging Face (MiniMaxAI/MiniMax-M2.1).
MiniMax-M2OpenCost-efficient agentic coding at scale on an Apache 2.0 open model.
MMiniMax
204K
$0.30
$1.20
API (model id: MiniMax-M2) via platform.minimax.io, plus open weights on Hugging Face and GitHub (Apache 2.0).
GLM-4.7OpenCost-conscious coding and agent tasks that still want deep reasoning.
ZZhipu AI (GLM)
200K
$0.60
$2.20
API (model id: glm-4.7) via Z.ai, plus open weights on Hugging Face (zai-org/GLM-4.7).
Kimi K2.7-CodeOpenAgentic coding, refactors, and tool-driven development inside IDEs and CLI agents.
KMoonshot AI (Kimi)
256K
$0.72
$3.50
API (model id: kimi-k2.7-code) plus open weights on Hugging Face under a Modified MIT license.
gpt-oss-120bOpenTeams that want to self-host a capable model.
OpenAI
128K
Open
Open
Open weights
gpt-oss-20bOpenOn-device and low-footprint deployments.
OpenAI
128K
Open
Open
Open weights
Qwen3-Coder-480B-A35BOpenAgentic coding, tool use and browser automation.
Alibaba (Qwen)
256K
Open
Open
API + open weights
Qwen3-Coder-NextOpenLighter self hosted coding agents.
Alibaba (Qwen)
256K
Open
Open
API + open weights
Qwen3-VLOpenMultimodal document and UI understanding.
Alibaba (Qwen)
256K
Open
Open
API + open weights
Qwen3.6-27BOpenStrong coding on a single accessible open model.
Alibaba (Qwen)
256K
Open
Open
Open weights
Qwen3-235B-A22BOpenSelf hosted general reasoning at scale.
Alibaba (Qwen)
256K
Open
Open
API + open weights
Kimi K2 (Instruct)OpenSelf-hosting and research on a widely adopted open MoE base.
KMoonshot AI (Kimi)
128K
Open
Open
API (model id: kimi-k2-0711-preview) plus open weights on Hugging Face (moonshotai/Kimi-K2-Instruct).
GLM-5OpenSelf-hosting and research on Zhipu's frontier open base.
ZZhipu AI (GLM)
1M
Open
Open
Open weights on Hugging Face (zai-org) plus API via Z.ai.
MiniMax-Text-01OpenExtreme long-context research and self-hosting on an open linear-attention model.
MMiniMax
4M
Open
Open
API (model id: MiniMax-Text-01) via platform.minimax.io, plus open weights on Hugging Face and GitHub.
MiniMax-VL-01OpenOpen multimodal research and self-hosting alongside MiniMax-Text-01.
MMiniMax
1M
Open
Open
Open weights on Hugging Face and GitHub (MiniMax-AI/MiniMax-01).
Hunyuan-LargeOpenSelf-hosting and research on a large open Tencent base.
TTencent (Hunyuan)
256K
Open
Open
Open weights on Hugging Face (tencent/Tencent-Hunyuan-Large) and GitHub, plus API via Tencent Cloud.
Hunyuan-A13BOpenLocal deployment and cost-sensitive inference on an open model.
TTencent (Hunyuan)
256K
Open
Open
Open weights on Hugging Face and GitHub (Tencent-Hunyuan), plus API via Tencent Cloud.
HunyuanImage 3.0OpenOpen text-to-image generation and research.
TTencent (Hunyuan)
N/A
Open
Open
Open weights on GitHub and Hugging Face (Tencent-Hunyuan/HunyuanImage-3.0), plus Tencent Cloud image APIs.
HunyuanVideoOpenOpen text-to-video and image-to-video generation and research.
TTencent (Hunyuan)
N/A
Open
Open
Open weights on Hugging Face (tencent/HunyuanVideo) and GitHub, plus Tencent Cloud video APIs.
Hunyuan3D 2.5OpenGame, product, and design workflows that need open 3D asset generation.
TTencent (Hunyuan)
N/A
Open
Open
Open weights on Hugging Face and GitHub (Tencent-Hunyuan), plus Tencent Cloud 3D APIs.
Command RLegacyCost sensitive RAG and chat.
CCohere
128K
$0.15
$0.60
API + open weights
GLM-4.6LegacySolid general coding and reasoning where you want a proven, cheaper checkpoint.
ZZhipu AI (GLM)
200K
$0.43
$1.74
API (model id: glm-4.6) via Z.ai, plus open weights on Hugging Face.
GLM-4.5LegacyGeneral coding and reasoning on a stable, widely supported checkpoint.
ZZhipu AI (GLM)
131K
$0.60
$2.20
API (model id: glm-4.5) via Z.ai, plus open weights on Hugging Face (zai-org/GLM-4.5).
GLM-4.5VLegacyMultimodal understanding on a proven, lower-cost checkpoint.
ZZhipu AI (GLM)
131K
$0.60
$1.20
API (model id: glm-4.5v) via Z.ai, plus open weights on Hugging Face.
Mistral Large 2LegacyWorkloads pinned to the older Large line.
Mistral AI
128K
$2.00
$6.00
API + open weights
Command R+LegacyExisting RAG stacks built on the R+ line.
CCohere
128K
$2.50
$10.00
API + open weights
Llama 3.3 70BLegacyTeams wanting a proven mid size open text model.
Meta
128K
Open
Open
Open weights + third-party API
Llama 3.2 90B VisionLegacyMultimodal apps that need self hosted image plus text.
Meta
128K
Open
Open
Open weights + third-party API
Llama 3.2 11B VisionLegacyLighter multimodal workloads and edge image tasks.
Meta
128K
Open
Open
Open weights + third-party API
Llama 3.2 3BLegacyEdge and mobile assistants where footprint matters.
Meta
128K
Open
Open
Open weights + third-party API
Llama 3.1 405BLegacyResearch and high end self hosted deployments.
Meta
128K
Open
Open
Open weights + third-party API
Llama 3.1 8BLegacyCost sensitive text tasks and fine tuning starters.
Meta
128K
Open
Open
Open weights + third-party API
DeepSeek-R1LegacySelf hosted reasoning and research on open weights.
DeepSeek
128K
Open
Open
Open weights
DeepSeek-V3.1LegacySelf hosting the earlier V3 line.
DeepSeek
128K
Open
Open
Open weights

Common questions

What are the best open-weight AI models?

The highest-tier open-weight models Vikshy tracks include Hunyuan Hy3 (Tencent (Hunyuan)), MiniMax-M3 (MiniMax), GLM-5.2 (Zhipu AI (GLM)). Open weights let you self-host, fine-tune, and run offline, at the cost of providing your own compute.

Is "open weights" the same as "open source"?

Not exactly. Open weights means the trained model file is downloadable; full open source would also include training data and code. Most "open" AI models are open-weight, under a range of licences.

← Back to the full model catalogue