Make
Choosing an LLM for computer use and agents
Ranked by the share of tasks finished by operating a real computer screen.
- Highest rated
- Claude Opus 5 $10
- On a budget
- MiniMax-M3 Open weights $0.53
- Gap in success rate: 26.8 points
- Worth 26.8 successes out of 100 tries
Success rate and price
The closer to the top left, the cheaper and better rated the model.
AnthropicOpenAIMiniMaxAlibabaMoonshot
Best-rated model at each price
First line: model / Success rate (%), second line: price estimate
| Model | Developer | Success rate (%) | Price |
|---|---|---|---|
| Claude Opus 5 | Anthropic | $10 (in $5, out $25) | |
| GPT-5.6 Sol | OpenAI | $8 (in $4, out $20) | |
| Claude Sonnet 4.6 | Anthropic | $6 (in $3, out $15) | |
| MiniMax-M3 | MiniMax | $0.53 (in $0.3, out $1.2) |
All models (highest success rate first)
Second line: price (input / output) / context / released / supports Tools Image JSON □ Compare
| Pick | Compare | Model | Developer | Success rate (%) | Terminal tasks (%) | Pro work (%) | Input $ /1M | Output $ /1M | Context | Supports | Weights | Released | Price and specs |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Claude Opus 5 | Anthropic | — | 65.8 | $5 | $25 | 1M | $5 / $25 1M 2026-07 | ||||||
| GPT-5.6 Sol | OpenAI | — | — | $4 | $20 | 1.05M | $4 / $20 1.05M 2026-07 | ||||||
| Claude Opus 4.8 | Anthropic | — | 48.9 | $5 | $25 | 1M | $5 / $25 1M 2026-05 | ||||||
| Claude Opus 4.7 | Anthropic | 80.2 | 49.2 | $5 | $25 | 1M | $5 / $25 1M 2026-04 | ||||||
| GPT-5.5 | OpenAI | 84.7 | 55.1 | $5 | $30 | 1.05M | $5 / $30 1.05M 2026-04 | ||||||
| Claude Sonnet 4.6 | Anthropic | 53.4 | 43.0 | $3 | $15 | 1M | $3 / $15 1M 2026-02 | ||||||
| Kimi K2.6 | Moonshot | — | — | $0.95 | $4 | 262.1K | $0.95 / $4 262.1K 2026-04 | ||||||
| MiniMax-M3 | MiniMax | — | 37.7 | $0.3 | $1.2 | 1.05M | $0.3 / $1.2 1.05M 2026-06 | ||||||
| Qwen3.7 Plus | Alibaba | — | — | $0.4 | $1.6 | 1M | $0.4 / $1.6 1M 2026-06 |
How to read this page
- Also shown (not used for ranking): success rates on terminal work and professional work tasks. Only the 9 models measured on this evaluation are listed (the newest was released in 2026-07).
- Bars are drawn from 0% to 40%.
- The success rate is the share of tries that succeeded. A gap of 1 point is 1 success out of 100 tries.
- Only some models have values for Terminal tasks.
- Only models rated higher than every cheaper model are listed, best first. For about the same rating, pick the cheapest row in this table. The top row within the price you can pay is the model with the highest success rate at that price.
- “On a budget” is the highest-rated model at $2 or less per million tokens (or at your price cap, when that is lower).
- Best at each price
- Prices are what developers pay to use a model through its API, separate from app subscriptions. The unit is one million tokens. A token is a small piece of text (about three quarters of an English word); one million tokens is roughly 8 to 10 novels. Reading (input) and writing (output) are priced differently; here they are blended 3 : 1. For example, at $10, one everyday question and answer (about 1,000 tokens) costs about 1 cent. Models without a known price are left out.
- In the Supports column the icons are, from left: can use tools / can read images / can answer in a fixed format (JSON); faded means not supported or unknown.
- Open-weights models have their contents (the weights) published, so you can also run them on your own computer or server. Models without the open-weights mark are normally used through the developer's own services (apps or API); for some models the weights status is unknown.
- Names such as ChatGPT next to a model are the consumer services its developer offers. Check on the service itself whether that model is available there.
- Vertical is Success rate, horizontal is price (the horizontal scale gets coarser to the right). Points joined by the step line are the best-rated models at each price. Hover or tap a dot to see the model's name.
Data: Arena, Epoch AI, Swallow LLM team (CC BY 4.0), Models.dev (MIT), updated 2026-10-24, data sources