---
title: "Open-Weight Model Rankings 2026"
canonical: "https://router.xark.io/model-rankings"
description: "Every open-weight model we serve, ranked by price, discount depth, context window and licence -- computed from our own rate table, not a borrowed benchmark."
section: "core"
updated: "2026-09-02"
source: "https://router.xark.io/model-rankings.md"
---

# Open-Weight Model Rankings 2026

Every open-weight model we serve, ranked by price, discount depth, context window and licence -- computed from our own rate table, not a borrowed benchmark.

We do not rank quality, reasoning or coding ability -- we have not benchmarked any model here. Every ranking below is a real property of the rate table: price, discount, context window, or licence family.

**Availability:** 7 of the 24 catalogued models are callable today. The rest are listed at their intended rate but no configured upstream serves them, so a request naming one returns `no available channel`. Every table below marks which is which.

## Agent-workload discount, ranked

| Model | Rate-card discount | Agent-workload discount |
| --- | --- | --- |
| Kimi K2.6 | 42% | 68% |
| GLM-5.2 | 42% | 68% |
| DeepSeek V4 Pro | 37% | 66% |
| Kimi K3 | 40% | 65% |
| DeepSeek V4 Flash | 36% | 65% |

## Context window, ranked widest to narrowest

| Model | Context window | Callable today |
| --- | --- | --- |
| Llama 4 Maverick | 1M | No |
| Kimi K3 | 512K | Yes |
| Kimi K2.6 | 256K | Yes |
| Qwen3 Max Instruct | 256K | No |
| Mistral Large 3 | 256K | No |
| GLM-5.2 | 200K | Yes |
| MiniMax M2 | 200K | No |
| DeepSeek V4 Pro | 160K | Yes |
| DeepSeek V4 Flash | 160K | Yes |
| GLM-5.2 Air | 128K | No |
| Qwen3 235B A22B | 128K | No |
| GPT-OSS 120B | 128K | No |
| Qwen3 Embedding 8B | 32K | Yes |
| BGE-M3 | 8K | No |

## Licence family distribution

| Family | Models |
| --- | --- |
| Apache 2.0 | 11 |
| MIT | 7 |
| Modified MIT | 2 |
| Llama 4 Community | 1 |
| LTXV Open Weights | 1 |
| Stability Community | 1 |
| Tencent Hunyuan Community | 1 |

Full reference: /licences. Interactive version with charts: /model-rankings.