Moonshot AI's Kimi K3 release is official as the largest open-source model to date.

Majority view

Most of the concrete, checkable parts of the image text about Moonshot AI’s Kimi K3 – that it is officially released, open‑weight, around 2.8T parameters, has a 1M‑token context window, native vision, and API pricing of about $3 per 1M input tokens and $15 per 1M output tokens – are supported by Moonshot’s own docs and multiple independent reports.[1][2][3][5][7][8][13] By contrast, statements that K3 is strictly “Claude Fable 5 and GPT‑5.6 Sol level,” that it is “the largest open‑source model to date,” and that analysts place its quality between GPT‑5.6 and Fable 5 are marketing/interpretive claims; current public benchmarks show K3 close to top proprietary models on some tasks but do not establish a firm consensus that it is second only to those systems or universally between them in quality.[1][2][5][13] Thus, the factual release, scale, context, modality, and pricing details are well‑supported, while the precise comparative performance rankings and “largest ever” superlatives are not fully validated and remain partly promotional.

Credible dissent

There is no dissent on basic facts like the existence of K3, its approximate parameter count, 1M context window, multimodal capabilities, and its $3/$15 API pricing; these points are repeated across Moonshot documentation, independent tech media, and API aggregators.[1][2][3][5][7][8][13] Substantial disagreement or at least uncertainty appears around the more ambitious comparative claims: different pre‑release and news sources report parameter counts between 2.5T and 3T, and some still describe K3 as upcoming or unreleased around mid‑2026.[4][11][12] Also, while several analyses say K3 “nears” GPT‑5.6 Sol and top models, they stop short of affirming it is definitively second only to Claude Fable 5 and GPT‑5.6 Sol or that it is clearly the largest open‑source model in existence, treating those as company claims rather than settled facts.[1][2][5][13]