Open-source models
Models with published weights on Hugging Face — inspect, self-host, or call them here. 161 of 338 catalog models match as of July 2026.
- 161
- Models
- $0.018
- Input from /1M
- 10M
- Max context
- Jun 17, 2026
- Newest release
161 open-source models are live on OpenKey right now, out of 329 total — priced anywhere from $0 to $3.00 per million output... input, actually, with input ranging $0 to $3.00/M across the set. The newest addition is North Mini Code (free), which landed 2026-06-17 with a 256K context window and a $0 price tag. At the other end, GLM 5.2 (2026-06-16) tops the design_arena 3d leaderboard at rank 1 (elo 1376) and runs $0.93/$3.00 per million in/out. If you want frontier open-weight quality you pay for it; if you want free, there are 25 free models on the platform and several show up in this exact list.
Top picks
GLM 5.2
Ranks #1 on design_arena 3d (elo 1376), the best benchmark result of any model in this list. It's also the newest top-tier release here (2026-06-16), with a 1,048,576-token context window and $0.93/$3.00 per million input/output — OpenKey price $0.9579/$3.09 after the 3% fee.
DeepSeek V4 Pro
Rank 7 on design_arena 3d (elo 1327) at $0.435/$0.87 per million tokens — roughly half GLM 5.2's price for a top-10 3d result. 1M context. Good pick if you want near-frontier design benchmarks without the premium rate.
Nemotron 3 Ultra (free)
Same model, same benchmark (design_arena 3d rank 40, elo 1213) as its paid twin, but $0 in/out. 1,000,000-token context. If budget is the constraint and you can tolerate a rank-40 result instead of top-10, this is free capacity, not free-tier gimmick.
All 161 models
Full catalog →North Mini Code (free)FreeCohere256KFreeFreectx 256K · inFreeoutFree
GLM 5.2Z.ai1.0Mctx 1.0M · inout
Kimi K2.7 CodeMoonshot AI262Kctx 262K · inout
Nex-N2-ProNex262Kctx 262K · inout
Nemotron 3 UltraNVIDIA1Mctx 1M · inout
Nemotron 3 Ultra (free)FreeNVIDIA1MFreeFreectx 1M · inFreeoutFree
Nemotron 3.5 Content Safety (free)FreeNVIDIA128KFreeFreectx 128K · inFreeoutFree
MiniMax M3MiniMax1.0Mctx 1.0M · inout
Step 3.7 FlashStepFun256Kctx 256K · inout
Granite 4.1 8BIBM131Kctx 131K · inout
Nemotron 3 Nano Omni (free)FreeNVIDIA256KFreeFreectx 256K · inFreeoutFree
Laguna M.1poolside262Kctx 262K · inout
Laguna M.1 (free)Freepoolside262KFreeFreectx 262K · inFreeoutFree
Laguna XS.2poolside262Kctx 262K · inout
Laguna XS.2 (free)Freepoolside262KFreeFreectx 262K · inFreeoutFree
Qwen3.6 27BQwen262Kctx 262K · inout
Qwen3.6 35B A3BQwen262Kctx 262K · inout
DeepSeek V4 FlashDeepSeek1.0Mctx 1.0M · inout
DeepSeek V4 ProDeepSeek1.0Mctx 1.0M · inout
Hy3 previewTencent262Kctx 262K · inout
MiMo-V2.5Xiaomi1.0Mctx 1.0M · inout
MiMo-V2.5-ProXiaomi1.0Mctx 1.0M · inout
Kimi K2.6Moonshot AI262Kctx 262K · inout
GLM 5.1Z.ai203Kctx 203K · inout
Gemma 4 26B A4B Google262Kctx 262K · inout
Gemma 4 26B A4B (free)FreeGoogle262KFreeFreectx 262K · inFreeoutFree
Gemma 4 31BGoogle262Kctx 262K · inout
Gemma 4 31B (free)FreeGoogle262KFreeFreectx 262K · inFreeoutFree
Trinity Large ThinkingArcee AI262Kctx 262K · inout
Reka EdgeReka16Kctx 16K · inout
MiniMax M2.7MiniMax205Kctx 205K · inout
Mistral Small 4Mistral AI262Kctx 262K · inout
Nemotron 3 SuperNVIDIA1Mctx 1M · inout
Nemotron 3 Super (free)FreeNVIDIA1MFreeFreectx 1M · inFreeoutFree
Qwen3.5-9BQwen262Kctx 262K · inout
LFM2-24B-A2BLiquid AI128Kctx 128K · inout
Qwen3.5-122B-A10BQwen262Kctx 262K · inout
Qwen3.5-27BQwen262Kctx 262K · inout
Qwen3.5-35B-A3BQwen262Kctx 262K · inout
Qwen3.5 397B A17BQwen256Kctx 256K · inout
MiniMax M2.5MiniMax205Kctx 205K · inout
GLM 5Z.ai203Kctx 203K · inout
Qwen3 Coder NextQwen262Kctx 262K · inout
Step 3.5 FlashStepFun262Kctx 262K · inout
Kimi K2.5Moonshot AI262Kctx 262K · inout
LFM2.5-1.2B-Instruct (free)FreeLiquid AI33KFreeFreectx 33K · inFreeoutFree
LFM2.5-1.2B-Thinking (free)FreeLiquid AI33KFreeFreectx 33K · inFreeoutFree
GLM 4.7 FlashZ.ai203Kctx 203K · inout
MiniMax M2.1MiniMax205Kctx 205K · inout
GLM 4.7Z.ai203Kctx 203K · inout
Nemotron 3 Nano 30B A3BNVIDIA262Kctx 262K · inout
Nemotron 3 Nano 30B A3B (free)FreeNVIDIA256KFreeFreectx 256K · inFreeoutFree
Devstral 2 2512Mistral AI262Kctx 262K · inout
GLM 4.6VZ.ai131Kctx 131K · inout
Ministral 3 14B 2512Mistral AI262Kctx 262K · inout
Ministral 3 3B 2512Mistral AI131Kctx 131K · inout
Ministral 3 8B 2512Mistral AI262Kctx 262K · inout
Trinity MiniArcee AI131Kctx 131K · inout
DeepSeek V3.2DeepSeek131Kctx 131K · inout
Olmo 3 32B ThinkAi266Kctx 66K · inout
Kimi K2 ThinkingMoonshot AI262Kctx 262K · inout
Voxtral Small 24B 2507Mistral AI32Kctx 32K · inout
gpt-oss-safeguard-20bOpenAI131Kctx 131K · inout
Nemotron Nano 12B 2 VL (free)FreeNVIDIA128KFreeFreectx 128K · inFreeoutFree
MiniMax M2MiniMax205Kctx 205K · inout
Qwen3 VL 32B InstructQwen262Kctx 262K · inout
Granite 4.0 MicroIBM131Kctx 131K · inout
Qwen3 VL 8B InstructQwen256Kctx 256K · inout
Qwen3 VL 8B ThinkingQwen256Kctx 256K · inout
Llama 3.3 Nemotron Super 49B V1.5NVIDIA131Kctx 131K · inout
Qwen3 VL 30B A3B InstructQwen262Kctx 262K · inout
Qwen3 VL 30B A3B ThinkingQwen131Kctx 131K · inout
GLM 4.6Z.ai203Kctx 203K · inout
DeepSeek V3.2 ExpDeepSeek164Kctx 164K · inout
Cydonia 24B V4.1TheDrummer131Kctx 131K · inout
Qwen3 VL 235B A22B InstructQwen262Kctx 262K · inout
Qwen3 VL 235B A22B ThinkingQwen131Kctx 131K · inout
DeepSeek V3.1 TerminusDeepSeek164Kctx 164K · inout
Qwen3 Next 80B A3B InstructQwen262Kctx 262K · inout
Qwen3 Next 80B A3B Instruct (free)FreeQwen262KFreeFreectx 262K · inFreeoutFree
Qwen3 Next 80B A3B ThinkingQwen262Kctx 262K · inout
Nemotron Nano 9B V2 (free)FreeNVIDIA128KFreeFreectx 128K · inFreeoutFree
Kimi K2 0905Moonshot AI262Kctx 262K · inout
Qwen3 30B A3B Thinking 2507Qwen131Kctx 131K · inout
Hermes 4 405BNous Research131Kctx 131K · inout
Hermes 4 70BNous Research131Kctx 131K · inout
DeepSeek V3.1DeepSeek164Kctx 164K · inout
GLM 4.5VZ.ai66Kctx 66K · inout
Jamba Large 1.7AI21 Labs256Kctx 256K · inout
gpt-oss-120bOpenAI131Kctx 131K · inout
gpt-oss-120b (free)FreeOpenAI131KFreeFreectx 131K · inFreeoutFree
gpt-oss-20bOpenAI131Kctx 131K · inout
gpt-oss-20b (free)FreeOpenAI131KFreeFreectx 131K · inFreeoutFree
Qwen3 Coder 30B A3B InstructQwen160Kctx 160K · inout
Qwen3 30B A3B Instruct 2507Qwen131Kctx 131K · inout
Qwen3 235B A22B Thinking 2507Qwen262Kctx 262K · inout
GLM 4.5Z.ai131Kctx 131K · inout
GLM 4.5 AirZ.ai131Kctx 131K · inout
Qwen3 Coder 480B A35BQwen1.0Mctx 1.0M · inout
Qwen3 Coder 480B A35B (free)FreeQwen1.0MFreeFreectx 1.0M · inFreeoutFree
UI-TARS 7B ByteDance128Kctx 128K · inout
Qwen3 235B A22B Instruct 2507Qwen262Kctx 262K · inout
Kimi K2 0711Moonshot AI131Kctx 131K · inout
Uncensored (free)FreeCognitive Computations33KFreeFreectx 33K · inFreeoutFree
Hunyuan A13B InstructTencent131Kctx 131K · inout
ERNIE 4.5 VL 424B A47B Baidu131Kctx 131K · inout
Mistral Small 3.2 24BMistral AI128Kctx 128K · inout
R1 0528DeepSeek164Kctx 164K · inout
Gemma 3n 4BGoogle33Kctx 33K · inout
Llama Guard 4 12BMeta AI164Kctx 164K · inout
Qwen3 14BQwen132Kctx 132K · inout
Qwen3 235B A22BQwen131Kctx 131K · inout
Qwen3 30B A3BQwen131Kctx 131K · inout
Qwen3 32BQwen131Kctx 131K · inout
Qwen3 8BQwen131Kctx 131K · inout
Llama 4 MaverickMeta AI1.0Mctx 1.0M · inout
Llama 4 ScoutMeta AI10Mctx 10M · inout
DeepSeek V3 0324DeepSeek164Kctx 164K · inout
Mistral Small 3.1 24BMistral AI128Kctx 128K · inout
Command ACohere256Kctx 256K · inout
Gemma 3 12BGoogle131Kctx 131K · inout
Gemma 3 4BGoogle131Kctx 131K · inout
Gemma 3 27BGoogle131Kctx 131K · inout
Reka Flash 3Reka66Kctx 66K · inout
Skyfall 36B V2TheDrummer33Kctx 33K · inout
Aion-1.0-MiniAion Labs131Kctx 131K · inout
Qwen2.5 VL 72B InstructQwen131Kctx 131K · inout
Mistral Small 3Mistral AI33Kctx 33K · inout
R1 Distill Llama 70BDeepSeek128Kctx 128K · inout
R1DeepSeek164Kctx 164K · inout
MiniMax-01MiniMax1.0Mctx 1.0M · inout
Phi 4Microsoft16Kctx 16K · inout
Llama 3.1 70B Hanami x1Sao10K16Kctx 16K · inout
DeepSeek V3DeepSeek131Kctx 131K · inout
Llama 3.3 Euryale 70BSao10K131Kctx 131K · inout
Llama 3.3 70B InstructMeta AI131Kctx 131K · inout
Llama 3.3 70B Instruct (free)FreeMeta AI131KFreeFreectx 131K · inFreeoutFree
Qwen2.5 Coder 32B InstructQwen128Kctx 128K · inout
UnslopNemo 12BTheDrummer33Kctx 33K · inout
Magnum v4 72BAnthracite33Kctx 33K · inout
Qwen2.5 7B InstructQwen131Kctx 131K · inout
Rocinante 12BTheDrummer33Kctx 33K · inout
Llama 3.2 11B Vision InstructMeta AI131Kctx 131K · inout
Llama 3.2 1B InstructMeta AI131Kctx 131K · inout
Llama 3.2 3B InstructMeta AI131Kctx 131K · inout
Llama 3.2 3B Instruct (free)FreeMeta AI131KFreeFreectx 131K · inFreeoutFree
Qwen2.5 72B InstructQwen131Kctx 131K · inout
Llama 3.1 Euryale 70B v2.2Sao10K131Kctx 131K · inout
Hermes 3 70B InstructNous Research131Kctx 131K · inout
Hermes 3 405B InstructNous Research131Kctx 131K · inout
Hermes 3 405B Instruct (free)FreeNous Research131KFreeFreectx 131K · inFreeoutFree
Llama 3 8B LunarisSao10K8Kctx 8K · inout
Llama 3.1 70B InstructMeta AI131Kctx 131K · inout
Llama 3.1 8B InstructMeta AI131Kctx 131K · inout
Mistral NemoMistral AI131Kctx 131K · inout
Gemma 2 27BGoogle8Kctx 8K · inout
Llama 3 8B InstructMeta AI8Kctx 8K · inout
Mixtral 8x22B InstructMistral AI66Kctx 66K · inout
WizardLM-2 8x22BMicrosoft66Kctx 66K · inout
ReMM SLERP 13BUndi956Kctx 6K · inout
MythoMax 13BGryphe4Kctx 4K · inout
Prices per 1M tokens, flat 3% fee included. Hover or tap a price for the math.
How to choose
First tradeoff: price per token vs. benchmark rank. GLM 5.2 sits at $0.93/$3.00 per million and rank 1 on design_arena 3d; DeepSeek V4 Flash is $0.09/$0.18 with a much lower rank (23) on the svg benchmark. Second: context window isn't free performance — GLM 5.2 and several Qwen3.5/Nemotron models offer 1M+ tokens, but a big window doesn't move the elo score. Third: free tiers (Nemotron 3 Ultra free, Laguna M.1 free, Gemma 4 free variants) usually mirror a paid model's exact benchmark — use them to test before committing spend, not as a permanent substitute if you need frontier rank.
Questions
- What's the newest open-source model on OpenKey?
- North Mini Code (free), added 2026-06-17. It's a text-to-text model with a 256,000-token context window and no input/output cost. No benchmark rank is listed for it yet.
- Which open-source model has the best benchmark score in this list?
- GLM 5.2, ranked #1 on design_arena 3d with an elo of 1376. It was created 2026-06-16, supports a 1,048,576-token context, and costs $0.93 per million input tokens and $3.00 per million output tokens before the OpenKey fee.
- Are there free open-source models with real benchmark results?
- Yes. Nemotron 3 Ultra (free) matches its paid counterpart's design_arena 3d rank 40 (elo 1213) at $0 cost, with a 1,000,000-token context window. OpenKey lists 25 free models total across the platform.