> How far open-weights AI models trail the closed frontier. Today the best open model, Kimi K3, matches what closed models first reached about 5 months earlier.

The gap your audience does not see

# Open vs closed

Open-weight models anyone can download and run trail the closed frontier by months, not years.

5months

open models behind today

closed models reached this level in Apr 2026

172

best measured open model

Kimi K3

74

open models above GPT-4 level

$0.036/M

cheapest open above GPT-4: gpt-oss-20b

## How far behind open models are · months, at each new open record

Each point is a new best open model, placed when its weights were published. Its height is how long before that a closed model had already reached the same score; zero means no closed model had that score yet. Points before the first measured closed model are left out. Hover for both models.

## Strongest open model from each lab · 12 labs · measured models

[Run it yourself →](https://themodelindex.org/local/)

| MODEL | LAB | REGION | INDEX SCORE | WEIGHTS OUT | CHEAPEST $/M | DOWNLOAD |
| --- | --- | --- | --- | --- | --- | --- |
| [Kimi K3](https://themodelindex.org/models/kimi-k3/) | Moonshot | China | 172 | Jul 2026 | $3.02 | [Weights ↗](https://huggingface.co/moonshotai/Kimi-K3) |
| [GLM-5.3](https://themodelindex.org/models/glm-5-3/) | Zhipu | China | 167 | Aug 2026 | $0.26 | [Weights ↗](https://huggingface.co/zai-org/GLM-5.3) |
| [DeepSeek-V4-Pro](https://themodelindex.org/models/deepseek-v4-pro-2026-08/) | DeepSeek | China | 167 | Aug 2026 | $1.98 | – |
| [TM Inkling-Small](https://themodelindex.org/models/inkling-small/) | Thinking Machines | United States | 154 | Jul 2026 | $0.64 | [Weights ↗](https://huggingface.co/thinkingmachines/Inkling-Small) |
| [Qwen3.8-27B](https://themodelindex.org/models/qwen3-8-27b/) | Alibaba | China | 152 | Aug 2026 | $0.58 | [Weights ↗](https://huggingface.co/Qwen/Qwen3.8-27B) |
| [MiniMax-M2.5](https://themodelindex.org/models/minimax-m2-5/) | MiniMax | China | 146 | Feb 2026 | $0.44 | [Weights ↗](https://huggingface.co/MiniMaxAI/MiniMax-M2.5) |
| [Nemotron 3 Ultra](https://themodelindex.org/models/nemotron-3-ultra/) | NVIDIA | United States | 145 | Jun 2026 | $0.93 | [Weights ↗](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16) |
| [Gemma 4](https://themodelindex.org/models/gemma-4/) | Google | United States | 138 | Apr 2026 | $0.15 | [Weights ↗](https://huggingface.co/google/gemma-4-31B-it) |
| [Mistral Medium 3.5](https://themodelindex.org/models/mistral-medium-3-5/) | Mistral | Europe | 134 | Apr 2026 | $3.0 | – |
| [gpt-oss-120b](https://themodelindex.org/models/gpt-oss-120b/) | OpenAI | United States | 131 | Aug 2025 | $0.065 | [Weights ↗](https://huggingface.co/openai/gpt-oss-120b) |
| [Llama 4 Maverick](https://themodelindex.org/models/llama-4-maverick/) | Meta | United States | 115 | Apr 2025 | $0.30 | [Weights ↗](https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct) |
| [Phi-4](https://themodelindex.org/models/phi-4/) | Microsoft | United States | 112 | Dec 2024 | $0.088 | [Weights ↗](https://huggingface.co/microsoft/phi-4) |

Weights out is when the weights were published, which can trail the announcement. Cheapest price is the lowest API provider on OpenRouter, blended 3:1 input to output.

---
Source: https://themodelindex.org/open/ · The Model Index · data as of 2026-10-01