model releaseMoonshot AI

Moonshot AI releases Kimi K3 with 2.7 trillion parameters, claims performance on par with Anthropic Fable 5

TL;DR

Moonshot AI released Kimi K3 on July 16, 2026, featuring 2.7 trillion parameters—the largest open-weight model to date. The company claims K3 performs competitively with Anthropic's Fable 5 while costing $15 per million output tokens compared to Fable's $50.

3 min read
1

Kimi K3 — Quick Specs

Context window1049K tokens
Input$3/1M tokens
Output$15/1M tokens

Moonshot AI releases Kimi K3 with 2.7 trillion parameters, claims performance on par with Anthropic Fable 5

Moonshot AI released Kimi K3 on July 16, 2026, featuring 2.7 trillion parameters—making it the largest open-weight large language model available. The Chinese startup claims K3 performs competitively with Anthropic's Fable 5, currently the most advanced widely available AI model.

Key specifications

  • Parameters: 2.7 trillion (compared to DeepSeek V4's 1.6 trillion)
  • Pricing: $15 per million output tokens
  • Model type: Open-weight coding model
  • Performance claims: Competitive with Fable 5, substantially outperforms Opus 4.8, GPT 5.6 Sol, and GPT 5.5 according to Moonshot

According to Moonshot AI, K3 "operates with minimal human oversight, can sustain long engineering sessions, navigate massive repositories, and orchestrate terminal tools." On the company's benchmarks, K3 consistently ranks within the top three models.

Pricing comparison

K3 costs $15 per million output tokens—expensive by Chinese standards but significantly cheaper than U.S. equivalents:

  • Anthropic Fable 5: $50 per million output tokens
  • z.ai GLM-5.2: $4.40 per million output tokens
  • DeepSeek V4: $0.87 per million output tokens

Existing market traction

Moonshot's previous models have already gained adoption among U.S. companies. Cursor used Kimi to help build Composer 2, its AI coding agent. DoorDash delegates "lower-level work to Kimi K2.6," according to CTO Andy Fang. Thinking Machines used Kimi K2.5 to generate post-training data for its Inkling model released July 15.

Context on Fable and Mythos

Anthroptic's Mythos 5 model, on which Fable 5 is based, is reportedly the most capable model for cyber-related tasks, but access is restricted to enterprises in Anthropic's Glasswing program for critical infrastructure security. The U.S. government temporarily imposed export controls on both Mythos and Fable after Amazon researchers jailbroke Fable's guardrails.

Analysts did not expect China to produce a Fable-level model until early 2027, making K3's release months ahead of projections.

Company background

Moonshot AI raised $2 billion in May 2026, valuing the company at over $20 billion. The company's annual recurring revenue exceeds $200 million, according to its financial advisor. Backers include Alibaba, Tencent, Meituan, and Hongshan Capital. Moonshot is reportedly preparing for an IPO in Hong Kong.

Policy implications

U.S. export controls barred Chinese developers from accessing advanced AI processors, forcing companies like Moonshot to focus on efficiency. "We knew we didn't have the luxury to simply scale up compute," said Moonshot AI president Yutong Zhang at the World Economic Forum earlier this year. "That forced us to focus on fundamental research and efficiency."

Anthroptic has accused Moonshot, z.ai, Minimax, Alibaba, and DeepSeek of "illicit" distillation attacks—using outputs from larger U.S. models to train smaller, more efficient models. U.S. politicians are considering measures to prevent such distillation and to curb the appeal of Chinese open-source models.

What this means

K3's release demonstrates that Chinese developers can build open-weight systems competitive with Anthropic's and OpenAI's flagship models despite U.S. export controls on AI chips. The model's earlier-than-expected arrival will likely intensify debates over U.S. AI policy—either prompting looser controls to help U.S. companies compete, or stricter measures to limit China's AI capabilities. For enterprises, K3 offers frontier-level performance at 70% lower cost than Fable 5, though as an open-weight model it requires more technical expertise and cloud infrastructure to deploy.

Related Articles

model release

Alibaba Releases Qwen3.8-Max, a 2.4 Trillion-Parameter Model Built for Multi-Day Autonomous Tasks

Alibaba has released Qwen3.8-Max, a 2.4-trillion-parameter model with 95 billion active parameters per query, designed to run autonomous tasks over multiple days. The company claims it hits 93 on PaperBench and rivals Claude Opus 4.8 and GPT-5.6 Sol on internal benchmarks, with open weights arriving next week.

model release

LG AI Research Releases K-EXAONE 2.0, a 750B-Parameter Open-Weight MoE Model with 262K Context

LG AI Research has released K-EXAONE 2.0, a 750-billion-parameter mixture-of-experts language model with 37B active parameters, a 262,144-token context window, and support for 10 languages. The model is open-weighted under Apache 2.0 and claims competitive results against Qwen3.5, GLM-5.1, and DeepSeek-V4 Pro on reasoning, coding, and long-context benchmarks.

model release

MiniMax Releases H3, a 33B-Parameter Omni-Modal Model That Generates 2K Video With Native Stereo Audio

MiniMax has published MiniMax-H3, a 33-billion-parameter omni-modal generative model capable of producing up to 15 seconds of 2K video with native stereo audio. The model accepts text, image, video, and audio inputs, though its full 2K pipeline depends on a hosted preprocessing component not included in the open-source release.

model release

Anthropic's Claude Opus 5 Generates Full 3D Games From a Single Text Prompt, No Assets Required

Anthropic's Claude Opus 5 can generate playable 3D games, including first-person shooters and Minecraft clones, from a single text prompt with zero external assets. Community tests claim it outperforms GPT-5.6 Sol and Kimi K3 in physics realism and mechanical complexity, though no standardized benchmark has confirmed the comparisons.

Comments

Loading...