Chinese AI Models Increase Price Pressure as Kimi K3 Opens Its Weights
ABC reported on August 1 Australian time that cheaper Chinese-developed models are changing how some businesses allocate AI workloads. Moonshot AI released its 2.8-trillion-parameter Kimi K3 on July 16 and published the weights on July 27; Moonshot says the model trails the leading proprietary systems overall, while independent reporting describes growing interest in lower-cost, self-hosted alternatives.
ABC reported on August 1 Australian time that increasingly capable Chinese-developed models are changing the economics of AI for some businesses. The report points to recent releases from Moonshot AI and Z.ai, alongside DeepSeek, as companies compare model quality, inference cost, deployment control, and jurisdictional risk.
Moonshot introduced Kimi K3 on July 16. Its official release describes a mixture-of-experts model with 2.8 trillion total parameters, native vision, and a one-million-token context window. Moonshot says K3 still trails Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol overall, while reporting competitive results on parts of its coding and agent evaluation suite. Those comparisons are vendor-reported and should not be treated as an independent ranking.
The company said at launch that the full weights would arrive by July 27. Moonshot's official GitHub repository now provides the model architecture, evaluation notes, deployment guidance, weights, and license. That changes the practical decision from choosing only between hosted APIs to deciding whether a team should also operate an open-weight model itself.
Cost is shifting workload placement
ABC interviewed Australian companies that are already changing how they route model traffic. Relevance AI said the share of its traffic sent to open-weight models had risen from roughly 5% to 7.5% at the start of 2026 to about 20% to 25%. Another engineering provider told ABC that some customers were seeking large savings as token bills grew, while advisers described routing difficult tasks to premium models and simpler work to cheaper alternatives. These are individual company reports, not market-wide adoption estimates.
Fortune separately reported substantial provider-price gaps. Its comparison put Kimi K3 at $15 per million output tokens, below Anthropic's cited $50 rate but above the cited prices for DeepSeek and Z.ai models. Pricing can change, and a lower API rate does not by itself establish equivalent quality, reliability, or total cost.
Open weights move costs rather than erase them
Self-hosting can reduce dependence on a single API provider and give teams more control over data location, customization, and capacity. It also transfers responsibility for GPUs, serving, observability, patching, access control, evaluation, and incident response.
For procurement teams, the useful comparison is therefore workload-specific: benchmark the exact tasks, measure end-to-end cost and latency, test safety and structured outputs, and review the model license and hosting jurisdiction. ABC's reporting also notes concerns about bias, trust, regulation, and software-supply-chain risk. The current competitive shift is real, but the least expensive token is not automatically the least expensive production system.
Key Points
- 1Moonshot released the 2.8-trillion-parameter Kimi K3 on July 16 and made its weights available on July 27, with performance claims still requiring independent workload testing.
- 2ABC documented individual Australian companies increasing open-weight use and routing simpler work to lower-cost models, but those examples are not market-wide adoption estimates.
- 3Open weights can reduce API dependence while shifting infrastructure, evaluation, security, licensing, and operational responsibility to the deploying organization.
Scoring Rationale
The story documents a meaningful change in model procurement: Chinese open-weight systems are becoming credible lower-cost workload options for some organizations. The impact is material for platform and finance teams, although vendor benchmarks, case-study adoption figures, and fast-changing prices require careful validation.
Sources
Primary source and supporting public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
