---
title: "Open-Source vs. Closed AI Models: How Big Is the Gap?"
date: 2026-10-06
url: https://modevc.com/charts/open-source-vs-closed-ai-models-gap/
keywords: ["llm benchmark comparison", "open source vs closed source llm", "best open source llm"]
publisher: Mode VC
---

# Open-Source vs. Closed AI Models: How Big Is the Gap?

The best open-weight model would win 44% of blind votes against the best closed one. The model is becoming a commodity input.

## Chart: The best open model trails the best closed model by 42 Arena points. In a blind vote it would still win 44% of the time, up from 31% at the widest gap.

Best proprietary rating minus best openly licensed rating on the Arena text board (overall), Arena points, Jan 2024 – 2 Oct 2026. Below 0 = an open model leads.

| Series | Date | Value | Note |
|---|---|---|---|
| Gap | 2024-01-09 | 67 | mixtral-8x7b-instruct-v0.1 |
| Gap | 2024-01-18 | 66 | mixtral-8x7b-instruct-v0.1 |
| Gap | 2024-01-25 | 66 | mixtral-8x7b-instruct-v0.1 |
| Gap | 2024-02-02 | 135 | mixtral-8x7b-instruct-v0.1 |
| Gap | 2024-02-15 | 107 | qwen1.5-72b-chat |
| Gap | 2024-03-05 | 104 | qwen1.5-72b-chat |
| Gap | 2024-03-07 | 105 | qwen1.5-72b-chat |
| Gap | 2024-03-13 | 104 | qwen1.5-72b-chat |
| Gap | 2024-03-26 | 105 | qwen1.5-72b-chat |
| Gap | 2024-03-27 | 105 | qwen1.5-72b-chat |
| Gap | 2024-04-03 | 105 | qwen1.5-72b-chat |
| Gap | 2024-04-09 | 62 | command-r-plus |
| Gap | 2024-04-10 | 61 | command-r-plus |
| Gap | 2024-04-11 | 68 | command-r-plus |
| Gap | 2024-04-13 | 67 | command-r-plus |
| Gap | 2024-04-18 | 66 | command-r-plus |
| Gap | 2024-04-19 | 60 | llama-3-70b-instruct |
| Gap | 2024-04-22 | 49 | llama-3-70b-instruct |
| Gap | 2024-04-26 | 50 | llama-3-70b-instruct |
| Gap | 2024-05-01 | 49 | llama-3-70b-instruct |
| Gap | 2024-05-08 | 50 | llama-3-70b-instruct |
| Gap | 2024-05-15 | 86 | llama-3-70b-instruct |
| Gap | 2024-05-16 | 85 | llama-3-70b-instruct |
| Gap | 2024-05-19 | 84 | llama-3-70b-instruct |
| Gap | 2024-05-20 | 84 | llama-3-70b-instruct |
| Gap | 2024-05-27 | 80 | llama-3-70b-instruct |
| Gap | 2024-06-02 | 80 | llama-3-70b-instruct |
| Gap | 2024-06-06 | 80 | llama-3-70b-instruct |
| Gap | 2024-06-11 | 80 | llama-3-70b-instruct |
| Gap | 2024-06-17 | 75 | nemotron-4-340b-instruct |
| Gap | 2024-06-21 | 77 | nemotron-4-340b-instruct |
| Gap | 2024-06-23 | 79 | nemotron-4-340b-instruct |
| Gap | 2024-06-26 | 80 | llama-3-70b-instruct |
| Gap | 2024-06-29 | 73 | gemma-2-27b-it |
| Gap | 2024-07-06 | 70 | gemma-2-27b-it |
| Gap | 2024-07-08 | 71 | gemma-2-27b-it |
| Gap | 2024-07-16 | 71 | gemma-2-27b-it |
| Gap | 2024-07-22 | 70 | gemma-2-27b-it |
| Gap | 2024-07-25 | 69 | gemma-2-27b-it |
| Gap | 2024-07-30 | 24 | llama-3.1-405b-instruct |
| Gap | 2024-07-31 | 24 | llama-3.1-405b-instruct |
| Gap | 2024-08-01 | 24 | llama-3.1-405b-instruct |
| Gap | 2024-08-05 | 23 | llama-3.1-405b-instruct |
| Gap | 2024-08-06 | 23 | llama-3.1-405b-instruct |
| Gap | 2024-08-13 | 22 | llama-3.1-405b-instruct |
| Gap | 2024-08-22 | 51 | llama-3.1-405b-instruct |
| Gap | 2024-08-23 | 51 | llama-3.1-405b-instruct |
| Gap | 2024-08-27 | 50 | llama-3.1-405b-instruct |
| Gap | 2024-08-28 | 50 | llama-3.1-405b-instruct |
| Gap | 2024-09-04 | 49 | llama-3.1-405b-instruct |
| Gap | 2024-09-15 | 28 | llama-3.1-405b-instruct-fp8 |
| Gap | 2024-09-17 | 89 | llama-3.1-405b-instruct-fp8 |
| Gap | 2024-09-27 | 73 | llama-3.1-405b-instruct-bf16 |
| Gap | 2024-10-07 | 68 | llama-3.1-405b-instruct-bf16 |
| Gap | 2024-10-15 | 68 | llama-3.1-405b-instruct-fp8 |
| Gap | 2024-10-23 | 70 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-10-28 | 64 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-11-04 | 63 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-11-12 | 64 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-11-13 | 64 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-11-20 | 65 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-11-21 | 65 | llama-3.1-nemotron-70b-instruct |
| Gap | 2024-11-22 | 60 | athene-v2-chat |
| Gap | 2024-12-01 | 57 | athene-v2-chat |
| Gap | 2024-12-05 | 57 | athene-v2-chat |
| Gap | 2024-12-10 | 56 | athene-v2-chat |
| Gap | 2024-12-15 | 57 | athene-v2-chat |
| Gap | 2024-12-18 | 58 | athene-v2-chat |
| Gap | 2024-12-22 | 57 | deepseek-v2.5-1210 |
| Gap | 2024-12-30 | 44 | deepseek-v3 |
| Gap | 2025-01-05 | 33 | deepseek-v3 |
| Gap | 2025-01-15 | 31 | deepseek-v3 |
| Gap | 2025-01-19 | 32 | deepseek-v3 |
| Gap | 2025-01-21 | 32 | deepseek-v3 |
| Gap | 2025-01-22 | 32 | deepseek-v3 |
| Gap | 2025-01-24 | -5 | deepseek-r1 |
| Gap | 2025-01-28 | -6 | deepseek-r1 |
| Gap | 2025-02-03 | -9 | deepseek-r1 |
| Gap | 2025-02-05 | -5 | deepseek-r1 |
| Gap | 2025-02-06 | -5 | deepseek-r1 |
| Gap | 2025-02-09 | -6 | deepseek-r1 |
| Gap | 2025-02-11 | -6 | deepseek-r1 |
| Gap | 2025-02-14 | -7 | deepseek-r1 |
| Gap | 2025-02-17 | -6 | deepseek-r1 |
| Gap | 2025-02-21 | -4 | deepseek-r1 |
| Gap | 2025-02-27 | -4 | deepseek-r1 |
| Gap | 2025-03-03 | 49 | deepseek-r1 |
| Gap | 2025-03-11 | 46 | deepseek-r1 |
| Gap | 2025-03-15 | 46 | deepseek-r1 |
| Gap | 2025-03-17 | 46 | deepseek-r1 |
| Gap | 2025-03-20 | 44 | deepseek-r1 |
| Gap | 2025-03-25 | 44 | deepseek-r1 |
| Gap | 2025-03-26 | 48 | deepseek-r1 |
| Gap | 2025-03-30 | 36 | deepseek-v3-0324 |
| Gap | 2025-04-05 | 40 | deepseek-v3-0324 |
| Gap | 2025-04-08 | 40 | deepseek-v3-0324 |
| Gap | 2025-04-09 | 36 | deepseek-v3-0324 |
| Gap | 2025-04-16 | 35 | deepseek-v3-0324 |
| Gap | 2025-04-23 | 45 | deepseek-v3-0324 |
| Gap | 2025-05-05 | 39 | deepseek-v3-0324 |
| Gap | 2025-05-06 | 39 | deepseek-v3-0324 |
| Gap | 2025-05-11 | 40 | deepseek-v3-0324 |
| Gap | 2025-05-19 | 41 | deepseek-v3-0324 |
| Gap | 2025-05-20 | 41 | deepseek-v3-0324 |
| Gap | 2025-05-22 | 41 | deepseek-v3-0324 |
| Gap | 2025-06-01 | 45 | deepseek-v3-0324 |
| Gap | 2025-06-03 | 44 | deepseek-v3-0324 |
| Gap | 2025-06-05 | 43 | deepseek-v3-0324 |
| Gap | 2025-06-11 | 43 | deepseek-v3-0324 |
| Gap | 2025-06-17 | 6 | deepseek-r1-0528 |
| Gap | 2025-06-18 | 6 | deepseek-r1-0528 |
| Gap | 2025-06-24 | 54 | deepseek-r1-0528 |
| Gap | 2025-07-01 | 49 | deepseek-r1-0528 |
| Gap | 2025-07-07 | 52 | deepseek-r1-0528 |
| Gap | 2025-07-15 | 50 | deepseek-r1-0528 |
| Gap | 2025-07-17 | 50 | deepseek-r1-0528 |
| Gap | 2025-07-25 | 48 | deepseek-r1-0528 |
| Gap | 2025-07-28 | 48 | deepseek-r1-0528 |
| Gap | 2025-08-01 | 39 | qwen3-235b-a22b-instruct-2507 |
| Gap | 2025-08-04 | 36 | glm-4.5 |
| Gap | 2025-08-07 | 38 | glm-4.5 |
| Gap | 2025-08-14 | 36 | glm-4.5 |
| Gap | 2025-08-18 | 37 | glm-4.5 |
| Gap | 2025-08-21 | 37 | glm-4.5 |
| Gap | 2025-08-27 | 37 | glm-4.5 |
| Gap | 2025-08-28 | 37 | glm-4.5 |
| Gap | 2025-09-08 | 37 | glm-4.5 |
| Gap | 2025-09-17 | 41 | glm-4.5 |
| Gap | 2025-09-18 | 41 | glm-4.5 |
| Gap | 2025-09-30 | 38 | qwen3-vl-235b-a22b-instruct |
| Gap | 2025-10-01 | 39 | glm-4.5 |
| Gap | 2025-10-03 | 23 | glm-4.6 |
| Gap | 2025-10-07 | 27 | glm-4.6 |
| Gap | 2025-10-16 | 26 | glm-4.6 |
| Gap | 2025-10-17 | 26 | glm-4.6 |
| Gap | 2025-11-05 | 21 | glm-4.6 |
| Gap | 2025-11-06 | 20 | glm-4.6 |
| Gap | 2025-11-09 | 21 | glm-4.6 |
| Gap | 2025-11-16 | 51 | glm-4.6 |
| Gap | 2025-11-17 | 53 | glm-4.6 |
| Gap | 2025-11-18 | 49 | glm-4.6 |
| Gap | 2025-11-20 | 48 | glm-4.6 |
| Gap | 2025-11-26 | 46 | glm-4.6 |
| Gap | 2025-12-01 | 45 | glm-4.6 |
| Gap | 2025-12-04 | 45 | glm-4.6 |
| Gap | 2025-12-05 | 45 | glm-4.6 |
| Gap | 2025-12-08 | 44 | glm-4.6 |
| Gap | 2025-12-11 | 46 | glm-4.6 |
| Gap | 2025-12-15 | 44 | glm-4.6 |
| Gap | 2025-12-16 | 45 | glm-4.6 |
| Gap | 2025-12-18 | 45 | glm-4.6 |
| Gap | 2025-12-19 | 44 | glm-4.6 |
| Gap | 2025-12-21 | 44 | glm-4.6 |
| Gap | 2025-12-23 | 44 | glm-4.6 |
| Gap | 2025-12-30 | 45 | glm-4.6 |
| Gap | 2026-01-09 | 44 | glm-4.6 |
| Gap | 2026-01-27 | 42 | glm-4.6 |
| Gap | 2026-01-29 | 35 | kimi-k2.5-thinking |
| Gap | 2026-02-06 | 57 | kimi-k2.5-thinking |
| Gap | 2026-02-09 | 60 | kimi-k2.5-thinking |
| Gap | 2026-02-10 | 41 | glm-5 |
| Gap | 2026-02-11 | 59 | kimi-k2.5-thinking |
| Gap | 2026-02-16 | 59 | glm-5 |
| Gap | 2026-02-17 | 58 | glm-5 |
| Gap | 2026-02-19 | 55 | qwen3.5-397b-a17b |
| Gap | 2026-02-20 | 54 | qwen3.5-397b-a17b |
| Gap | 2026-02-21 | 55 | qwen3.5-397b-a17b |
| Gap | 2026-02-22 | 55 | qwen3.5-397b-a17b |
| Gap | 2026-02-24 | 54 | kimi-k2.5-thinking |
| Gap | 2026-02-25 | 54 | qwen3.5-397b-a17b |
| Gap | 2026-02-26 | 50 | qwen3.5-397b-a17b |
| Gap | 2026-03-03 | 54 | qwen3.5-397b-a17b |
| Gap | 2026-03-04 | 55 | qwen3.5-397b-a17b |
| Gap | 2026-03-05 | 53 | qwen3.5-397b-a17b |
| Gap | 2026-03-06 | 54 | qwen3.5-397b-a17b |
| Gap | 2026-03-11 | 53 | kimi-k2.5-thinking |
| Gap | 2026-03-16 | 50 | kimi-k2.5-thinking |
| Gap | 2026-03-18 | 49 | kimi-k2.5-thinking |
| Gap | 2026-03-19 | 49 | kimi-k2.5-thinking |
| Gap | 2026-03-20 | 50 | kimi-k2.5-thinking |
| Gap | 2026-03-26 | 51 | kimi-k2.5-thinking |
| Gap | 2026-03-31 | 51 | kimi-k2.5-thinking |
| Gap | 2026-04-02 | 25 | gpt-5.4-high |
| Gap | 2026-04-07 | 25 | gpt-5.4-high |
| Gap | 2026-04-09 | 26 | gpt-5.4-high |
| Gap | 2026-04-10 | 26 | gpt-5.4-high |
| Gap | 2026-04-14 | 28 | gpt-5.4-high |
| Gap | 2026-04-17 | 28 | gpt-5.4-high |
| Gap | 2026-04-20 | 26 | gpt-5.4-high |
| Gap | 2026-04-22 | 35 | glm-5.1 |
| Gap | 2026-04-23 | 34 | glm-5.1 |
| Gap | 2026-04-27 | 32 | glm-5.1 |
| Gap | 2026-04-28 | 32 | glm-5.1 |
| Gap | 2026-04-29 | 34 | glm-5.1 |
| Gap | 2026-04-30 | 33 | glm-5.1 |
| Gap | 2026-05-01 | 32 | glm-5.1 |
| Gap | 2026-05-06 | 31 | glm-5.1 |
| Gap | 2026-05-07 | 32 | glm-5.1 |
| Gap | 2026-05-12 | 32 | glm-5.1 |
| Gap | 2026-05-14 | 30 | glm-5.1 |
| Gap | 2026-05-18 | 31 | glm-5.1 |
| Gap | 2026-05-19 | 31 | glm-5.1 |
| Gap | 2026-05-27 | 30 | glm-5.1 |
| Gap | 2026-06-03 | 30 | glm-5.1 |
| Gap | 2026-06-04 | 30 | glm-5.1 |
| Gap | 2026-06-05 | 31 | glm-5.1 |
| Gap | 2026-06-10 | 31 | glm-5.1 |
| Gap | 2026-06-16 | 31 | glm-5.1 |
| Gap | 2026-06-25 | 32 | glm-5.1 |
| Gap | 2026-07-02 | 34 | glm-5.1 |
| Gap | 2026-07-10 | 34 | glm-5.1 |
| Gap | 2026-07-12 | 34 | glm-5.1 |
| Gap | 2026-07-14 | 34 | glm-5.1 |
| Gap | 2026-07-16 | 35 | glm-5.1 |
| Gap | 2026-07-20 | 36 | glm-5.1 |
| Gap | 2026-07-21 | 37 | glm-5.1 |
| Gap | 2026-07-26 | 33 | kimi-k3-max |
| Gap | 2026-07-27 | 38 | kimi-k3-max |
| Gap | 2026-07-30 | 38 | kimi-k3-max |
| Gap | 2026-08-02 | 34 | kimi-k3-max |
| Gap | 2026-08-03 | 42 | glm-5.1 |
| Gap | 2026-08-06 | 32 | kimi-k3-max |
| Gap | 2026-08-10 | 32 | kimi-k3-max |
| Gap | 2026-08-11 | 32 | kimi-k3-max |
| Gap | 2026-08-12 | 31 | kimi-k3-max |
| Gap | 2026-08-19 | 28 | kimi-k3-max |
| Gap | 2026-08-21 | 27 | glm-5.3-max |
| Gap | 2026-08-27 | 29 | kimi-k3-max |
| Gap | 2026-09-01 | 29 | kimi-k3-max |
| Gap | 2026-09-02 | 38 | kimi-k3-max |
| Gap | 2026-09-11 | 33 | glm-5.3-max |
| Gap | 2026-09-13 | 32 | glm-5.3-max |
| Gap | 2026-09-25 | 27 | mimo-v2.6-pro |
| Gap | 2026-09-30 | 43 | mimo-v2.6-pro |
| Gap | 2026-10-02 | 42 | mimo-v2.6-pro |

Note: Win rate uses Arena's rating scale: P(open wins) = 1 / (1 + 10^(gap/400)). "Open" means any non-proprietary licence in the dataset, including non-commercial ones; it is not the OSI definition.

## The signal

- **Launches:** [Mistral Large 4](https://mistral.ai/news/mistral-large-4/) (6 Oct): 1-trillion-parameter multimodal open-weight-to-be model aimed at coding, agents and enterprise. Weight release was not confirmed when we checked.
- **Launch feed:** Of 60 named model launches since July, 28 state their weights: 25 closed, only 3 open.
- **Arena:** mimo-v2.6-pro is the top open model on 2 Oct 2026; gemini-4-argon-high is #1 overall.

## The Mode read

Mistral announced Large 4 on 6 October and describes it as an open-weight model. Most launches still keep weights closed, but the open models that do ship are close to the frontier.

On the Arena, **mimo-v2.6-pro** is now the best openly licensed model. It is 42 points behind gemini-4-argon-high. On Arena's rating scale, that gap means the open model wins about **44%** of blind head-to-head votes. At the widest gap since 2024 (2 Feb 2024) it would have won 31%. In February 2025, deepseek-r1 briefly led outright.

When the free option is this close, model access becomes a commodity input. For app builders this is good: switching cost falls, and so does the price of "good enough". For a startup whose whole product is a thin layer on one closed model, it is a warning.

> When the free model wins 44% of the time, the model is not the moat.

**What we would back:** Teams that turn cheap, open intelligence into a product people already pay for: vertical, embedded, hard to switch away from.

## Sources

- [Arena leaderboard dataset (lmarena-ai, CC BY 4.0)](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset)
- [Mistral Large 4 announcement](https://mistral.ai/news/mistral-large-4/)
- [Lab posts + launch trackers (llmgateway, thursdai)](https://thursdai.news)