Opus 5 or Kimi K3 Max: 18 points apart, 40 percent cheaper
Kimi is 40 percent cheaper and only 18 points lower on the web-dev leaderboard, the closest pair in our table. What the extra 40 percent buys is not a better model; it is official tooling and an Iran column that has at least been measured.
Specifications where both sides have a verified number
| Claude Opus 5 | Kimi K3 Max | |
|---|---|---|
| API model ID | claude-opus-5 | kimi-k3 |
| Context window | 1 million tokens | 1 million tokens |
| Input price | $5 per million tokens | $3 per million tokens |
| Output price | $25 per million tokens | $15 per million tokens |
A row where one side has no verified number is left out entirely, because readers read a half-empty row in favour of the filled side.
The closest pair in this table
Opus 5 scores 1692 on WebDev Arena and Kimi K3 Max scores 1674. Eighteen points, on a leaderboard where the distance from top to floor is 138. Meanwhile Opus 5 charges $25 per million output tokens and Kimi charges $15. If those were the only two numbers you had, Kimi would be the obvious pick.
The first row of the table above carries a point
Two rows in that table need reading carefully. The context window row prints one million tokens on both sides, because the Kimi figure is 1,048,576 and our formatter rounds it; the real difference is under five percent and makes nothing possible or impossible in practice. The second row matters more: in the API id row, the Kimi side reads kimi-k3 rather than kimi-k3-max. No model called Kimi K3 Max exists in the Moonshot catalogue; that name is a leaderboard row describing a higher reasoning effort. What you buy is kimi-k3, and the price and window belong to it.
What the 40 percent actually pays for
In the tooling maturity column Opus 5 takes 4 of 4 and Kimi takes 1 of 4: the Moonshot platform documentation carries neither an official command line tool nor an editor extension. The second is the Iran column. Anthropic publishes a supported-countries list and Iran is not on it, so Opus 5 is recorded as blocked; Moonshot publishes no list at all, so Kimi stays unchecked. In our ranking a vendor that documents its restrictions outranks one that documents nothing.
Pick this if Claude Opus 5
- You need official tooling: a command line, an editor extension, batch execution, presence on a major cloud.
- You want your country access status documented at all, even when the answer is no.
- Those eighteen points decide it for you, because your work sits at the hard edge.
Pick this if Kimi K3 Max
- The invoice decides: $3 and $15 against $5 and $25.
- Your team writes its own tooling layer and needs no ready-made SDK.
- You send very long inputs: the Kimi window is 1,048,576 tokens, slightly above one million.
Questions people actually ask
Is Kimi K3 Max different from Kimi K3
No. The Moonshot catalogue holds only kimi-k3, and the max suffix is a reasoning effort level that appears in the leaderboard row. The price and window in the table above belong to that base model.
Which should I take for code
If your team builds its own tooling and cost matters, Kimi; eighteen points is felt less in practice than a forty percent price gap. If you are relying on official tooling and support, Opus 5, and it is first in our table too.