Kimi K2 Thinking model launches, touted to outperform GPT-5 via API
I just tried a brand-new language model that popped up on a public API, and the curl snippet below shows exactly how you can talk to it. The buzz says Kimi K2 Thinking might outpace GPT-5, but the real proof will be in the request you send and the reply you get. The endpoint lives at api.moonshot.cn, a domain that hints at a Chinese-based service, and it expects a plain OpenAI-style payload.
In the example, the headers are simple: Content-Type set to application/json and an Authorization token placeholder. The JSON body names the model, sends a tiny “hello” prompt, and sets temperature to 1.0. If you’ve used REST calls before, this layout will look familiar; if not, the snippet is a concrete entry point.
For anyone wanting to test the performance claims or hook the model into an app, this is probably the first line of code you’ll run. The next part spells out the exact request format, so you can see what’s required.
model parameter:$ curl https://api.moonshot.cn/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $MOONSHOT_API_KEY" \ -d '{ "model": "kimi-k2-thinking", "messages": [ {"role": "user", "content": "hello"} ], "temperature": 1.0 }' For more details on API use, checkout this guide. Also Read: Kimi OK Computer: A Hands-On Guide to the Free AI Agent Prompt: "Simulate a structured debate between Nikola Tesla and Thomas Edison on the ethics of AI today. Ground their arguments in their actual writings, then extend their worldviews to comment on issues like deepfakes, automation, and open-source models." Output: My Take: Kimi K2 Thinking delivered an outstanding performance on the task of simulating a historically grounded debate between Nikola Tesla and Thomas Edison on the ethics of modern AI.
Does Kimi K2 Thinking actually beat GPT-5? The piece says yes, pointing to a slide made from one prompt and the model’s ability to browse the web on its own. Still, that claim leans on a single author’s taste rather than any solid benchmark, so the comparison stays shaky.
Kimi K2 Thinking is an open-source thinking agent; it can grow its reasoning steps at test time and pull in tools on the fly, which sounds like it could handle a range of problems. Its API is already live - a quick curl call using the model name “kimi-k2-thinking” and a user message shows it works right away. Because the team hasn’t revealed the model’s parameter count or training data, it’s hard to gauge how deep its understanding really is.
The talk of “test-time scaling” also lacks numbers, so we can’t tell if it consistently outperforms other models across the board. For developers who like an extensible, self-acting system, it might be worth a look, but the evidence so far doesn’t conclusively prove it’s better than existing LLMs. Independent tests would be needed to back up the article’s upbeat take.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI OpenAI says AI saves knowledge workers 40‑80 minutes; use yields five‑fold gains
LLMs & Generative AI Grok Chat: AI for debugging, building, testing web apps with voice and images
LLMs & Generative AI Samsung adds Vision AI Companion, an AI Bixby, to TVs for real‑time queries
Related Topic Developers say Sora, unlike Vine/TikTok, is not about people in social media
Research & Benchmarks
Related Topic Jony Ive and Sam Altman unveil AI hardware prototype; OpenAI plans launch under two years
Business & Startups
Common Questions Answered
What endpoint and payload format does the Kimi K2 Thinking API use?
The Kimi K2 Thinking model is accessed via the endpoint https://api.moonshot.cn/v1/chat/completions and expects an OpenAI‑style JSON payload. The request must include a Content‑Type header of application/json, an Authorization Bearer token, and fields such as "model", "messages", and "temperature".
How does the article claim Kimi K2 Thinking outperforms GPT‑5?
The article asserts that Kimi K2 Thinking can surpass GPT‑5 by generating slide decks from a single prompt and performing autonomous web‑browsing, showcasing more advanced reasoning steps. However, it notes that these claims rely on a single author's preference rather than systematic benchmark testing.
What role does the "temperature" parameter play in the Kimi K2 Thinking API request shown?
In the curl example, the "temperature" parameter is set to 1.0, which instructs the model to produce more diverse and creative outputs. Higher temperature values increase randomness, while lower values make responses more deterministic.
What capabilities does Kimi K2 Thinking’s open‑source design provide at test time?
Kimi K2 Thinking’s open‑source architecture expands its reasoning steps dynamically during inference and can integrate external tools such as web browsers on the fly. This design aims to enable flexible problem‑solving and more complex, multi‑step tasks compared to static models.