Rankings

You no longer need a $15-per-million flagship for most work. These are the best cheap AI models in 2026: everything here costs at most about $1 per million input tokens, ranked by how much real capability you keep while the price drops.

Hermes runs all day, so the model price is your real subscription fee. These are the best cheap models for Hermes right now: everything costs at most about $0.75 per million input tokens, ranked by how much agent capability survives the price cut.

Chinese labs now ship frontier-class models at a fraction of US flagship prices, and most of them publish open weights. These are the best Chinese AI models in 2026, ranked on real agentic-coding performance and value, from Moonshot, DeepSeek, Alibaba, MiniMax, Z.AI, and Xiaomi.

Want a coding model that costs $0? OpenRouter serves several capable coding models for free (rate limits apply). These are the best free coding models on OpenRouter, ranked on real agentic-coding ability, from quick edits to full agent runs.

Want to run Hermes without paying per token? OpenRouter serves plenty of capable models at $0 (rate limits apply). These are the best free models for Hermes, grouped by what you run it for: general agent work, coding, and fast high-volume steps.

Want to run OpenClaw without paying per token? OpenRouter serves plenty of capable models at $0 (rate limits apply). These are the best free models for OpenClaw, grouped by what you run it for: general operator work, coding, and fast high-volume steps.

OpenRouter lists dozens of models you can call at $0 (rate limits apply). These are the best free models on OpenRouter right now, grouped by what you want to do: general use, coding, chat and reasoning, vision, and uncensored.

Claude Code is tuned for the Claude family, but it's worth knowing where each model lands on capability and cost for your workload.

Codex only runs OpenAI models, so the real choice is which GPT tier fits your work. Here's how the three options in the picker compare on capability and cost.

Hermes runs your skills locally and leans on the model for planning and skill use. These are the models that pair best with it right now, grouped by what you actually want to spend and ranked on real agentic-coding performance.

OpenClaw lives in your messaging apps and acts on your behalf, so it rewards models with reliable skill use, long sessions, and sane costs. These are the models that pair best with it right now, grouped by what you actually want to spend and weighed against the always-on token bill.

Running models locally with Ollama means no API keys, no rate limits, and no data leaving your machine. These are the best open-weight models to run locally in 2026, grouped by the hardware you actually have, from a 16GB laptop to a multi-GPU server.

Open-weight models you can download, inspect, and self-host now sit within reach of the closed frontier. These are the best open-source AI models in 2026, ranked on real capability, from full-size flagships to models you can run on a workstation.

Open-weight models you can inspect, fine-tune, and self-host. Ideal for privacy-sensitive or air-gapped Hermes setups. We rank a deep bench here because Hermes routes more than 380 different models in the wild, and most of its real volume goes to open weights.

Open-weight models you can inspect, fine-tune, and self-host. A natural fit for OpenClaw, which is a self-hostable operator runtime, so you can keep the whole stack in-house with no per-token bill. We rank a deep bench here because most of OpenClaw's real volume goes to open weights.

Alibaba's Qwen family is the broadest model lineup in open AI: frontier flagships, coding specialists, vision models, and tiny MoEs that run on a laptop. These are the best Qwen models in 2026 and what each one is actually for.