Kimi vs ChatGPT: Kimi now scores higher on reasoning, ChatGPT costs a third as much per task and won our 30 language test

Which is better, Kimi or ChatGPT?

Kimi leads on the score, but ChatGPT's Sol model is the better buy for what you pay. Artificial Analysis runs every model through the same ten tests and prices each task at the maker's list rates. Moonshot released its new model on 23 October 2026, and two days later it ranked fourth of 215 models. That's three points above the Sol model in ChatGPT Plus, for 2.9 times the cost of a task. You pay a lot for those three points.

Both makers charge $2.00 per million tokens in and $10.00 out for the two cheaper flagships. So why does Kimi cost more per task? It's because it writes a lot more of them. Kimi's new model used 55,000 output tokens on an average task, and the Sol model used 38,000, by the same Artificial Analysis count.

ChatGPT's models also guess less. AA-Omniscience rewards a right answer and takes points off a wrong one, so a model that says "I don't know" scores higher than one that bluffs. ChatGPT's Sol model scores 42 there, and Kimi's newest model scores 30. If you came here for Kimi K3 against ChatGPT, it's the bottom row: 44 on the index, four points under the older GPT-6 Sol.

Is Kimi better than ChatGPT for coding?

Kimi's newest model is, on the coding tests Artificial Analysis runs. It scores 67% on Terminal-Bench 4.0, which grades an agent working in a real terminal, to 56% for ChatGPT's Sol model and 59% for GPT-6 Astra. It also leads AutomationBench-AA, 73% to 65%. The older Kimi K3 scored 13% on Terminal-Bench 4.0 against 44% for GPT-6 Sol, so the jump in October is large.

It's also quicker to start. Through the API at maximum effort, Kimi's new model sent its first token after 3.07 seconds, and the Sol model took 236 seconds to start. A full task still took Kimi longer, at 846 seconds to 628, because it thinks for more tokens. ChatGPT runs its own settings in the app, so treat these as API numbers.

There's one place ChatGPT still wins on hard questions. On Humanity's Last Exam, the Sol model scores 53% and GPT-6 Astra 55%, against 52% for Kimi. On CritPt, a physics research test, the Sol model leads 32% to 25%. For other picks by coding task, see ChatGPT alternatives for coding.

Is Kimi or ChatGPT better in other languages?

ChatGPT is, by a wide margin, on Whizi's own test. Our Multilingual LLM Bench asks 150 questions we wrote from scratch, five in each of 30 languages, and grades every answer against a sourced key. We ran each model once at low reasoning and finished on 2 October 2026. Kimi's newest model came out after that run, so the Kimi row here is Kimi K3.

GPT-6 Sol beat Kimi K3 in 24 of the 30 languages and tied the other six. It lost none. Kimi K3 didn't even win in Chinese, where it got two of the five questions right and GPT-6 Sol got four of them. Five questions per language is a small sample, so the ranking of any one language is only a hint. The 32 point gap overall is four times the margin of error. It isn't close.

The free tier tells the same story. GPT-6 Luna, the model ChatGPT Free runs on, scored 44.7, or 13.4 points above Kimi K3, and it cost a cent for all 150 answers. Kimi K3 beat Luna in 5 languages and lost in 15.

Does ChatGPT know more recent events than Kimi?

Yes, by about five months. Whizi's Knowledge Cutoff Bench asks four dated questions for every month since January 2024, each keyed to an official record of something that happened that month. A model's measured cutoff is the last month where it still gets at least half of them right, averaged over three months. We ran it on 15 October 2026.

Kimi K3's memory ends in November 2025, five months before GPT-6 Sol's. That's a long time in AI. Moonshot doesn't publish a cutoff for it, and Kimi K3 was honest about the gap: it said it didn't know on 95% of the later questions, instead of making something up. Both apps can search the web, which covers news either model never saw.

How much do Kimi and ChatGPT cost?

Both are free to start. After that, ChatGPT's cheapest paid plan costs less than half of what Kimi's does. On the US App Store, read 25 October 2026, the Kimi app sells memberships from $19.00 a month. OpenAI sells ChatGPT Go for $8 a month in the US and ChatGPT Plus for $20.

OpenAI's pricing page lists the newer Sol model and GPT-6 Astra from Plus up, and neither on Free or Go. Moonshot's App Store listing says its newest model "is live" in the app. Its store page doesn't say which membership unlocks how much of it, so check inside the app before you pay.

On the API, the two of them cost the same for each token. As of October 2026, Moonshot's price list charges $2.00 per million input tokens and $10.00 per million output tokens for its newest model, and $3.00 and $15.00 for Kimi K3. OpenAI's price list charges $2.00 and $10.00 for the newer Sol model and $10.00 and $50.00 for GPT-6 Astra. Cached input costs $0.20 a million on Kimi and $0.10 on the Sol model.

Is Kimi open source, unlike ChatGPT?

Kimi K3 is, and Kimi's newest model isn't yet, though Moonshot says it will be. Moonshot published the Kimi K3 weights on Hugging Face in July 2026, and its list of models there ends with Kimi K3, read 25 October 2026. Artificial Analysis lists the new model as proprietary for now and says "the model weights are not publicly available". Moonshot's own guide to the new model calls it open weight and says "The full model weights will be released by Nov 4." So check Hugging Face after 4 November 2026. ChatGPT's models have never been downloadable.

The open one comes with a few conditions, though. The Kimi K3 License starts from MIT style wording and adds two rules. A company that sells the model as a service needs a separate deal with Moonshot once its group revenue passes $20 million over 12 months. And above 100 million monthly users or $20 million in monthly revenue, "Kimi K3" must be "prominently displayed on the user interface". Open source LLMs guide covers what it takes to run weights like these yourself.

Where do Kimi and ChatGPT keep your chats?

Kimi keeps them in China, and ChatGPT keeps them in the United States. The Kimi privacy policy for kimi.com and the apps, updated 24 August 2026, names Beijing Moonshot Technology and says it stores your personal information inside the People's Republic of China. OpenAI's privacy policy describes "processing and storing your Personal Data in our facilities and servers in the United States", and in other countries where its partners work.

Both may train on what you type. Only ChatGPT gives you a switch for it. In ChatGPT, open Settings, then Data controls, and turn off "Improve the model for everyone", per OpenAI's data controls FAQ. Kimi says it uses content only after strict de-identification, and to opt out you contact its customer service, which checks your identity first. Is DeepSeek safe walks through the same questions for the other big Chinese app.

Should you pick Kimi or ChatGPT?

Pick Kimi's newest model for coding agents and the top reasoning score. Pick ChatGPT for cost per task, other languages, recent events and fewer wrong guesses.

Whizi, which publishes this page, has Kimi K3 and ChatGPT's newer Sol model on its Pro plan at $29.99 a month, and the side by side view sends one prompt to both. Credits per message vary by model and can change, so pricing has the current rates. Where it loses: Pro reaches Kimi K3 and Kimi K2.5 only, so Kimi's newest model isn't on the plan named here. Its model page says which plan has it. The Kimi and GPT docs list every row.