Open-Weight • Self-Hostable • Buildable
Open DAO LLM Matrix
Overview
Key Finding: DeepSeek-V4-Flash delivers frontier agentic performance at ~$0.05–0.24/M on DeepInfra/OpenRouter with full MIT license and easy self-host paths. GLM-5.2 leads raw planning quality. Gemma-4-31B and Qwen3-32B are the easiest to run locally on consumer/prosumer hardware. All models below have permissive or community-friendly licenses. Kimi K3 (2.8T) joins as the largest open-weight model once weights drop on July 27.
Full Pricing Matrix (Hosted)
Input / Output Pricing per 1M Tokens — July 2026 (model × provider)
| Model | DeepInfra | OpenRouter | Moonshot | License | Context | Notes |
|---|---|---|---|---|---|---|
| DeepSeek-V4-Flash | $0.09 / $0.18 | $0.054 / $0.242 | — | MIT | 1M | Best price |
| DeepSeek-V4-Pro | $1.30 / $2.60 | $0.45 / $3.31 | — | MIT | 1M | Premium |
| GLM-5.2 | $0.45 / $3.10 | $0.45 / $3.31 | — | Apache-2.0 | 256k | Planning |
| Qwen3-235B-A22B | $0.09 / $0.55 | $0.20 / $0.88 | — | Apache-2.0 | 256k | Value leader |
| Llama-4-Scout | $0.10 / $0.30 | $0.12 / $0.35 | — | Llama 4 | 320k | Fast MoE |
| Gemma-4-31B | $0.13 / $0.38 | — | — | Apache-2.0 | 256k | Easy local |
| Nemotron-3-Ultra | $0.42 / $2.61 | $0.42 / $2.61 | — | OpenMDW | 1M | U.S. stack |
| Kimi K3 | — | $0.475 / $15.00* | $3.00 / $15.00 ($0.30 cache) |
Open (Jul 27) | 1M | Frontier 2.8T MoE |
* = effective price with high cache hit rate. Empty = provider does not offer this model. Data extracted 2026-07-25.