Two dollars per million input tokens. That number alone explains why developers worldwide sat up on October 24, 2026. Moonshot AI, a Chinese AI company, just released Kimi K3.1 Pro — a model that not only competes at the top of the benchmarks, but radically undercuts the competition on price. Anyone who has grumbled about the cost of large language models over the past few months should take a close look.
Kimi K3.1 Pro is available immediately through the Kimi API, Kimi.ai, Kimi Work, and Kimi Code. The model goes head-to-head with established names like Claude Opus 5.5 and GPT-6 Astra — at a fraction of the usual cost.
Table of Contents
Key Takeaways
- Kimi K3.1 Pro costs $2.00 per million input tokens and $10.00 per million output tokens, a third cheaper than its predecessor Kimi K3.
- The model has 2.8 trillion parameters total, with 104 billion active, and a 1-million-token context window.
- On the Artificial Analysis Intelligence Index, Kimi K3.1 Pro scores 55 points and lands in 4th place, just behind Claude Opus 5.5 and Claude Sonnet 5.5.
- Open weights are expected to be released by November 4, 2026, which would make it the strongest open AI model in the world.
- Moonshot AI deliberately throttled offensive cyber capabilities to reduce the risk of misuse.
What exactly is Kimi K3.1 Pro
Kimi K3.1 Pro is the latest language model from the Chinese company Moonshot AI. One important clarification up front: Moonshot AI has no connection to OpenAI, Anthropic, or Google. It is an independent Chinese provider that has built a strong reputation in large language models over the past year.
According to Artificial Analysis, Kimi K3.1 Pro has 2.8 trillion parameters in total, of which 104 billion are active at any given time. This Mixture-of-Experts architecture activates only part of the model for a given request — the result is lower compute cost alongside high model capacity.
The context window is 1 million tokens. That means Kimi K3.1 Pro can process extremely long documents, codebases, or conversation histories in a single pass without losing context.
Kimi K3.1 Pro pricing in detail
The most striking aspect of this release is arguably the pricing structure. Moonshot AI is deliberately positioning Kimi K3.1 Pro as an aggressively priced offering against Western competitors.
| Model variant | Input (per 1M tokens) | Output (per 1M tokens) | Cached input |
|---|---|---|---|
| Kimi K3 (predecessor) | $3.00 | $15.00 | $0.30 |
| Kimi K3.1 Pro (Standard) | $2.00 | $10.00 | $0.20 |
| Kimi K3.1 Pro (Highspeed) | $4.00 | $20.00 | $0.20 |
Compared to its predecessor Kimi K3, the standard variant of Kimi K3.1 Pro is roughly a third cheaper. For developers who need fast response times, Moonshot also offers a Highspeed variant, which delivers two to three times faster output according to the company — at double the price of the standard version. The cached-input price stays identical across both variants at $0.20 per million tokens.
"We aim not to release open weights that would extend the absolute frontier of cyber capability.", Moonshot AI
Kimi K3.1 Pro benchmark performance
Low prices alone rarely win developers over. What matters is how Kimi K3.1 Pro holds up against the competition. Two key benchmarks provide solid data here.
Terminal-Bench 4.0
On Terminal-Bench 4.0, a test of practical coding and terminal tasks, Kimi K3.1 Pro scores 65.7% according to Moonshot’s own figures — just behind Claude Opus 5.5 at 66.4%. GPT-6 Astra trails noticeably at 58.2%. For a model at this price point, that is a remarkable result.
Artificial Analysis Intelligence Index
The Artificial Analysis Intelligence Index, a broader benchmark of general model intelligence, shows a similar picture. Here Kimi K3.1 Pro scores 55 points and ranks 4th overall.
| Model | Intelligence Index |
|---|---|
| Claude Opus 5.5 | 58 |
| Claude Sonnet 5.5 | 56 |
| Muse Wonder Full | 56 |
| Kimi K3.1 Pro | 55 |
| GPT-6 Astra | 53 |
| Claude Fable 5.1 | 53 |
| Gemini 4 Argon | 53 |
Kimi K3.1 Pro sits just behind the three top-ranked models, but outperforms GPT-6 Astra, Claude Fable 5.1, and Gemini 4 Argon. For a model priced well below many of its rivals, that is a strong position in the upper tier of the absolute frontier.
Open weights: Kimi K3.1 Pro as the world’s strongest open model
Developers and researchers are watching one point especially closely: Moonshot AI plans to release the full model weights of Kimi K3.1 Pro by November 4, 2026. Open weights mean the model won’t just be usable through a paid API — research institutions, companies, and individual developers will be able to self-host and fine-tune it. Earlier in October, Moonshot already released Kimi K2.7 Code, a specialized coding model that K3.1 Pro builds directly upon.
If that timeline holds, Kimi K3.1 Pro would become the strongest open-weight model in the world based on currently available benchmark data — a significant moment for the open-source AI community, which has often lagged behind proprietary models.
Safety measures: throttled cyber capabilities
One detail that often gets lost in coverage deserves special attention. Moonshot AI says it deliberately took steps to limit Kimi K3.1 Pro’s offensive cyber capabilities. Specifically, training data was filtered to prevent the model from being misused as a tool for advanced cyberattacks.
On the internal xdev-bench-hard benchmark, which tests exploit capabilities, the released model scores 81.0%. An unfiltered internal checkpoint reportedly would have scored 90.5%. That roughly 9.5-point gap shows Moonshot deliberately sacrificed capability to reduce risk.
This self-restraint is especially relevant given the upcoming open-weight release. Once the weights are freely available, the model can no longer be centrally controlled. Moonshot frames the decision directly:
"We aim not to release open weights that would extend the absolute frontier of cyber capability."
What this means for developers
For developers and companies integrating AI models into their products, several practical considerations emerge:
- Cost efficiency: Lower token prices can translate into noticeable savings at high API volume compared to Western providers.
- Data privacy and compliance: Since Moonshot AI is a Chinese company, organizations should review which data protection rules and legal frameworks apply when using the API, particularly regarding regulations like the EU’s GDPR.
- Open-weight option: With the planned release of model weights, self-hosting becomes possible starting November 2026, offering more control over data.
- Speed versus cost: The Highspeed variant is worth it mainly for latency-sensitive applications, while the standard variant should be sufficient for most use cases.
These considerations are not a substitute for individual legal or technical advice, but offer an initial orientation for evaluating the model.
Conclusion
Kimi K3.1 Pro marks a notable step in the global competition among large language models. Moonshot AI, a Chinese company with no ties to OpenAI, Anthropic, or Google, shows with this release that strong performance and aggressive pricing don’t have to be mutually exclusive. With 65.7% on Terminal-Bench 4.0 and 4th place on the Artificial Analysis Intelligence Index, Kimi K3.1 Pro competes at the top tier while costing significantly less than many rival products.
The upcoming open-weight release by November 4, 2026 is likely to further shape the open-source AI landscape. At the same time, the deliberate throttling of offensive cyber capabilities shows that Moonshot AI is aware of the risks that come with open models.
Developers and companies considering Kimi K3.1 Pro should first evaluate the API in a limited test environment, weigh the cost structure against their own use cases, and closely follow developments around the open-weight release in November.
