Moonshot AI's trillion-parameter model undercuts Anthropic on price while chasing near-parity on coding benchmarks
The AI coding wars just got a new price leader. Moonshot AI’s Kimi 2.7, also known as K2.7 Code, is positioning itself as the budget-friendly alternative to Anthropic’s Claude Fable 5, one of the most capable coding models on the market. The pitch is straightforward: get close to the same performance for a fraction of the cost.
Kimi 2.7’s API pricing lands at $0.95 per million input tokens and $4.00 per million output tokens, with cache hits running just $0.19 per million tokens.
What Kimi 2.7 brings to the table #
Under the hood, Kimi 2.7 runs on a Mixture-of-Experts architecture. The model boasts up to one trillion total parameters, but only 32 billion are active at any given time. This design lets it punch above its weight on performance while keeping computational costs manageable.
The model’s core selling points are long-context reliability and higher task success rates. For developers working on real-world software engineering problems, where files are long and dependencies are tangled, that reliability matters more than raw benchmark scores on toy problems.
Moonshot AI, the Beijing-based lab behind the K2 model series, has been steadily climbing the ranks in the frontier model race. Kimi 2.7 represents their most aggressive move yet into the coding-specific AI space, a domain increasingly dominated by Western labs like Anthropic, OpenAI, and Google DeepMind.
The model it’s chasing #
Claude Fable 5, released on June 9, 2026, set a high bar. Anthropic’s latest model broke through the 90% threshold on core analytics benchmarks, a milestone that represents a significant jump in software engineering and analytical task performance.
Moonshot AI is betting that most developers don’t need the absolute best model. If Kimi 2.7 can deliver competitive coding capability at a price that doesn’t require a budget review meeting every quarter, the math starts to look very attractive for startups, independent developers, and mid-size engineering teams.
The bigger picture for AI competition #
The trillion-parameter scale of Kimi 2.7’s full architecture puts it in the same weight class as the largest models from any lab globally. The Mixture-of-Experts approach, which activates only 32 billion parameters per query, is what makes it economically viable to offer at sub-dollar input pricing.
Claude Fable 5 remains the benchmark for raw coding performance, particularly on complex analytical tasks where its 90%-plus scores speak for themselves. But Kimi 2.7 represents a viable alternative for teams where token budgets are tight and workloads are high-volume.
The cache hit pricing at $0.19 per million tokens is particularly noteworthy for production applications. Developers who structure their workflows to maximize cache reuse could see effective costs drop dramatically, making Kimi 2.7 one of the cheapest frontier-class coding models available for repetitive or templated tasks.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our