China’s Moonshot Prices Kimi at Half Opus’s Standard Rate

BEIJING, Oct. 24 — MoonshotMoonshot AI, a Chinese start-up, has released Kimi K3.1 Pro at half the standard price Anthropic lists for Claude Opus 5.5: $2 for every million tokens a customer sends the model and $10 for every million it generates, against $4 and $20. It says it will publish the model’s full weights, the trained parameters others need to run it on their own machines, by Nov. 4.

Artificial Analysis, an independent testing firm, says that when the weights are published K3.1 Pro will be the highest-scoring open-weights model on its Intelligence Index, narrowing the gap between the leading open-weights and proprietary models from 12 points to 3.

OpenRouter, a marketplace for AI models, listed it on Friday afternoon, New York time, early Saturday in Beijing. The new price is a third below the $3 and $15 Moonshot charges for Kimi K3, released in July.The ModelK3.1 Pro is a post-trained version of Kimi K3. Artificial Analysis lists it at 2.8 trillion parameters, 104 billion of them active; Moonshot calls it “3T-class.” Its context window is 1 million tokens. For American labs that sell their strongest models by the token, that makes it a rival near the top of the rankings and cheaper than Opus, one that others may soon be able to host for themselves, on terms Moonshot has not yet published.

The first benchmark Moonshot cites is Terminal-Bench 4.0, which sets coding agents to work on tasks in a computer terminal. It reports 65.7 percent for K3.1 Pro, against 66.4 percent for Opus 5.5 and 58.2 percent for OpenAI’s GPT-6 Astra, taking the rivals’ scores from their makers’ posts or the test’s public leaderboard, whichever was higher. Artificial Analysis, running the test itself, scored K3.1 Pro at 67 percent, “the highest score of any model.”

On the firm’s broader index, K3.1 Pro scored 55, fourth, behind Opus 5.5 at 58 and Anthropic’s Claude Sonnet 5.5 and Meta’s Muse Wonder Full at 56. Sonnet 5.5 lists at the same $2 and $10, so against it Kimi offers no discount; cached input costs 20 cents per million tokens at Kimi and Opus alike; and Anthropic’s batch service, for work that can wait, charges $2 and $10 for Opus. Moonshot’s faster version, K3.1 Pro Highspeed, which it says runs two to three times as fast, costs $4 and $20: Opus’s standard rate, and half Anthropic’s price for Opus in its fast mode. DeepSeek and Xiaomi sell cheaper models still, VentureBeat noted.

Moonshot says it held one capability back on purpose. For the open release it removed cybersecurity material from two stages of training, mid-training and reinforcement learning; on its in-house exploit test, a version trained without those cuts solved 90.5 percent of problems in three tries, and the final model 81.0 percent. K3.1 Pro’s cyber capability, the company wrote, “remains behind that of frontier proprietary models that have been publicly available for at least three months.”

The launch has also drawn allegations of undisclosed promotion. VentureBeat reported that an X user had posted screenshots of what he described as instructions for a Kimi promotional campaign, telling creators not to use the paid-partnership label; citing Pekingnology, it said the document “has not been independently authenticated,” that it was unclear whether it came from Moonshot, an agency or someone else, and that it had identified no confirmed company response.

The weights are not out yet, and Moonshot’s post does not say what license will govern them. Until then, customers reach the model through Moonshot’s own apps and API; the API platform’s privacy policy says “our servers are situated in Singapore.” The test comes by Nov. 4: whether the weights arrive, on what terms, and whether outside testers running the model themselves match Moonshot’s numbers.

Daily Cortex is written with AI models, including Anthropic’s Claude, and fact-checked with OpenAI’s GPT models.