Kimi K3: The Largest Open AI Model Yet, Claimed to Match Opus 4.8 at Sonnet 5 Pricing

In July 2026, the Chinese lab Moonshot AI released Kimi K3, an open language model with 2.8 trillion parameters and a Mixture of Experts architecture, of which about 50 billion are active per token. It is the largest open model ever published.

What Is Kimi K3

Kimi K3 is Moonshot AI's flagship open model, built on a mixture-of-experts (MoE) design. The 2.8T-A50B label means 2.8 trillion parameters in total and about 50 billion active at each generation step. This design gives the model the "knowledge" of a giant network, while its per-request compute cost behaves more like that of a dense model with ~50 billion parameters.

Key specifications from the announcement:

• Name: Kimi K3, developed by Moonshot AI (the creators of the Kimi line)
• Architecture: Mixture of Experts, 2.8 trillion parameters in total, ~50 billion active (notation 2.8T-A50B)
• Status: the largest open model at the time of release
• Quality: claimed to match Anthropic's Claude Opus 4.8
• Price: comparable to Claude Sonnet 5
• Predecessor: the open model Kimi K2

Opus 4.8 Quality at Sonnet 5 Pricing

Kimi K3 is positioned as a Claude Opus 4.8-class model in terms of quality, priced on par with Claude Sonnet 5, so flagship results at a mid-tier price. This price-to-quality ratio is what Latent Space put in the headline of its AINews newsletter breakdown.

For the market, this is a direct challenge to closed flagships. If an open-weights model really reaches top Opus-level performance on a range of tasks, teams get the option to deploy it on their own infrastructure, without paying per token and without sending data to an external API.

"It continues to be a great week for open models," says the AINews newsletter issue from

Latent Space.

Why It Matters for the Market

The Kimi K3 release extends a wave of strong open-weights launches. In recent cycles, Chinese teams have set the pace in the open-model race (Moonshot with its Kimi line, along with DeepSeek and Qwen), and each new flagship has narrowed the gap with the closed systems from OpenAI, Anthropic, and Google.

At 2.8 trillion parameters, the race moves into a new weight class: public weights of this size had simply never been released before. One important caveat: at launch, "Opus 4.8 class" remains a positioning claim, not an independently verified benchmark result. Real numbers will have to come from community testing.

What This Means

Open models have come close to the flagship level of the closed labs, and they do so at a mid-tier price. For developers and companies, this widens the choice: top-tier quality is now available not only through the APIs of the major vendors but also on their own servers.

Frequently Asked Questions

Who developed Kimi K3?

Kimi K3 was created by the Chinese AI lab Moonshot AI, the same team that released the earlier open models in the Kimi line, including Kimi K2.

What does 2.8T-A50B mean?

It describes the Mixture of Experts architecture: 2.8T is the model's total parameter count (2.8 trillion), and A50B is the number of parameters active per token (about 50 billion). Only a subset of the "experts" runs at each step, so the model computes faster than a dense network of the same size.