Kimi K3 is live on API and apps today, with full weights due July 27. Another open frontier option — and another reason to update your model map.
Via Kimi / Moonshot AI: Kimi K3: Open Frontier Intelligence
Moonshot AI introduced Kimi K3 on July 16, 2026: a 2.8-trillion-parameter model with native vision and a 1-million-token context window. The company calls it the world's first open 3T-class model, aimed at long-horizon coding, knowledge work, and reasoning.
Availability today: Kimi.com, Kimi Work, Kimi Code, and the Kimi API. Full model weights are scheduled for July 27, 2026. Moonshot is frank that overall performance still trails Claude Fable 5 and GPT 5.6 Sol — while claiming frontier-level results against the rest of what they tested.
For an owner-led firm, this is another Model Selection event: a large open option with published pricing and a near-term weight drop. It expands the menu. It does not replace a job map or a fallback.
Moonshot positions K3 as the latest step after months of pushing the open-model size frontier. Size is a signal. Continuity and fit are the decision.
Kimi K3 uses Kimi Delta Attention and Attention Residuals, with MoE sparsity that activates 16 of 896 experts under Stable LatentMoE. Moonshot says those changes yield about a 2.5× improvement in scaling efficiency versus Kimi K2. At launch, thinking effort defaults to max; lower-effort modes come later.
API pricing on the official platform: $0.30 per million tokens for cache-hit input, $3.00 for cache-miss input, and $15.00 for output — flat across context length. Moonshot cites cache hit rates above 90% in coding workloads on its Mooncake inference stack. That is useful if your firm runs long coding or agent sessions; it is still a bill that needs a ceiling.
The capability demos lean hard on long-horizon agentic work — kernel optimization, building a MiniTriton-like GPU compiler, chip design proofs, research pipelines, and vision-in-the-loop game/frontend iteration. Impressive for builders. For law, dental, medical, CPA, and RIA shops, the decision remains: which jobs justify this class of model, where do tokens run, and what happens if API access or weight distribution changes after July 27.
Open weights on a schedule are not the same as local deployment on day one. Until the checkpoint lands and your stack can host or route it through a trusted partner, treat K3 as another vendor option with a continuity plan — not a default for client data.
Kimi Enterprise is pitched for team privacy and member management with separation between personal and organization accounts — a reminder that product packaging and data controls still matter more than parameter counts when client files are involved.
Add Kimi K3 to the evaluation list. Do not make it the whole policy.
Also note the jurisdiction angle neutrally: Moonshot is a Beijing-based lab. For regulated US SMEs, that is a factor to weigh in residency and vendor-risk reviews — not a reason to ignore the model, and not a reason to adopt it without a written rule.
Kimi K3 is a 2.8T-parameter model... the world's first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning.
This release maps to Model Selection & Continuity Planning. The question is not "is Kimi K3 the best model?" It is "which model fits which job, at what cost and residency, with what fallback when the next 3T-class drop lands?"
AgentsROI stays vendor-neutral. I help owner-led firms in law, accounting, health, and wealth choose deliberately — cloud, hybrid, or local — and keep continuity so a headline open release does not become an accidental production dependency. If your team is already chasing every new frontier API, a Shadow-AI Risk Assessment or Workflow ROI Audit finds the gap between policy and practice.
Primary next step: update your model map for K3's API-now / weights-later timeline — before someone wires it into a client workflow on vibes alone.
Moonshot released Kimi K3 as an open frontier bet: huge scale, long context, native vision, API today, weights by July 27. Useful. The useful move for an SME owner is still the boring one: assign models to jobs, cap spend, and keep a fallback.
Ready to pressure-test your model map? Start with a Model Selection & Continuity review at AgentsROI.ai.
This article summarizes publicly reported information and is for general informational purposes only. It does not constitute legal, tax, financial, investment, security, or compliance advice. AgentsROI.ai is not a law firm, accounting firm, or registered investment adviser. Facts, pricing, statistics, and product capabilities cited here reflect the sources listed at the time of writing and may change. Readers should verify current information independently and consult qualified professionals regarding obligations specific to their industry, jurisdiction, and circumstances—including applicable New York State and New York City requirements. AgentsROI.ai may have commercial relationships with vendors mentioned; where material, such relationships are disclosed. Nothing in this article is an endorsement of any specific AI product, model, or provider.
