Chinese AI start-up Moonshot has unveiled the Kimi K3 large language model with one million context tokens. The model is built to compete with Anthropic’s Opus 4.8 per competitive coding benchmarks found in early tests. Benchmark leaks suggest Kimi K3 performance targets Anthropic’s Claude Fable5 head-on. Industry watchers say the release heightens the AI competition among Chinese and Western developers. Moonshot presents the model as a step towards narrowing the performance gap against leading global alternatives.