The AI coding wars simply acquired a brand new worth chief. Moonshot AI’s Kimi 2.7, also called K2.7 Code, is positioning itself because the budget-friendly various to Anthropic’s Claude Fable 5, one of the succesful coding fashions available on the market. The pitch is simple: get near the identical efficiency for a fraction of the price.
Kimi 2.7’s API pricing lands at $0.95 per million enter tokens and $4.00 per million output tokens, with cache hits working simply $0.19 per million tokens.
What Kimi 2.7 brings to the desk
Below the hood, Kimi 2.7 runs on a Combination-of-Consultants structure. The mannequin boasts as much as one trillion whole parameters, however solely 32 billion are lively at any given time. This design lets it punch above its weight on efficiency whereas maintaining computational prices manageable.
The mannequin’s core promoting factors are long-context reliability and better job success charges. For builders engaged on real-world software program engineering issues, the place recordsdata are lengthy and dependencies are tangled, that reliability issues greater than uncooked benchmark scores on toy issues.
Moonshot AI, the Beijing-based lab behind the K2 mannequin collection, has been steadily climbing the ranks within the frontier mannequin race. Kimi 2.7 represents their most aggressive transfer but into the coding-specific AI area, a website more and more dominated by Western labs like Anthropic, OpenAI, and Google DeepMind.
The mannequin it’s chasing
Claude Fable 5, launched on June 9, 2026, set a excessive bar. Anthropic’s newest mannequin broke by means of the 90% threshold on core analytics benchmarks, a milestone that represents a big bounce in software program engineering and analytical job efficiency.
Moonshot AI is betting that the majority builders don’t want the very best mannequin. If Kimi 2.7 can ship aggressive coding functionality at a worth that doesn’t require a funds assessment assembly each quarter, the maths begins to look very engaging for startups, unbiased builders, and mid-size engineering groups.
The larger image for AI competitors
The trillion-parameter scale of Kimi 2.7’s full structure places it in the identical weight class as the biggest fashions from any lab globally. The Combination-of-Consultants strategy, which prompts solely 32 billion parameters per question, is what makes it economically viable to supply at sub-dollar enter pricing.
Claude Fable 5 stays the benchmark for uncooked coding efficiency, notably on complicated analytical duties the place its 90%-plus scores converse for themselves. However Kimi 2.7 represents a viable various for groups the place token budgets are tight and workloads are high-volume.
The cache hit pricing at $0.19 per million tokens is especially noteworthy for manufacturing functions. Builders who construction their workflows to maximise cache reuse might see efficient prices drop dramatically, making Kimi 2.7 one of many least expensive frontier-class coding fashions accessible for repetitive or templated duties.