And the model wars continue to be interesting.
The recent debut of Grok 4 by xAI has sparked significant interest as it appears to compete effectively with leading AI models from OpenAI, Anthropic, and Google. Initial user feedback indicates that Grok 4 has outperformed its rivals in complex tasks, providing more comprehensive solutions than previous models. For example, Tim Hua, a former economist at Walmart, reported that Grok 4 successfully addressed three out of four parts of a demand estimation problem, surpassing other models that only managed one or two parts. However, despite its advancements, Grok 4 was released without the safety reports and testing protocols that its competitors prioritize, raising concerns about safety standards in the rapidly evolving AI landscape.
Yet even more interesting is Moonshot AI has launched Kimi K2, an open-source language model that outperforms the proprietary systems of OpenAI and Anthropic in several key benchmarks, particularly in coding and autonomous tasks. The model boasts one trillion total parameters and has achieved a remarkable 65.8 percent accuracy on the SWE-bench Verified software engineering benchmark, surpassing its competitors. Kimi K2 is designed to autonomously execute tasks, reducing the need for human intervention. The model’s optimization for agentic capabilities allows it to complete complex workflows effectively, with a striking 97.4 percent accuracy on mathematical reasoning tests compared to OpenAI’s GPT-4.1, which scored 92.4 percent. Kimi K2 is positioned as a more efficient and cost-effective solution compared to existing models, with Moonshot AI’s innovative MuonClip optimizer enabling stable training without significant computational overhead. The company’s pricing strategy further undercuts major players, charging only $0.15 per million input tokens for cache hits and $2.50 for output tokens.
Some companies are beginning to cancel their subscriptions to artificial intelligence chatbots like ChatGPT as they seek more cost-effective options. According to new data from payments startup Ramp, the percentage of corporate customers spending on AI products, including ChatGPT, slightly decreased from 42.5% to 42% between May and June 2025. Ramp’s economist, Ara Kharazian, noted that while the trend of canceling chatbot subscriptions is evident, overall corporate spending on AI continues to rise. Companies are increasingly turning to free alternatives, such as Google’s Gemini chatbot, which is now included by default in Workspace productivity applications, and Microsoft’s Copilot features, which are available at no cost to business users. Despite these cancellations, OpenAI reported significant growth in revenue from corporate subscriptions, adding one million paying business customers since February 2025.
Why do we care?
The model wars are entertaining, but they’re no longer where the money is. IT service providers should see this as the inflection point to pivot from being consumers of AI models to builders of AI-enhanced services.
I want to dwell on Kimi here for the value. Moonshot AI offers two tracks. A low barrier entry API. You pay $0.15 per million input tokens (cache hits) and $2.50 per million output tokens. SMBs and enterprises can quickly test and deploy AI functionality without investing in infrastructure or expertise. This lowers adoption friction—IT teams don’t need specialized ML engineers to get started.
As needs grow or API costs scale uncomfortably, businesses can download and self-host Kimi K2. This allows complete control over data privacy, model tuning, and system integration. It also enables fine-tuning for industry-specific tasks, which is where vertical differentiation happens.
That approach undermines the closed source options – they can’t simply price match. Now, self hosting isn’t easy, and lack SLA’s and other negatives too. But this is commoditizing the model market. My guidance:
Don’t sell access to AI models—sell business outcomes powered by AI orchestration.
Focus on vertical specialization, helping clients integrate commoditized LLMs into workflows that align with their unique industry needs.
Position as AI cost and governance advisors, ensuring clients get the right mix of open source, bundled, and premium models without overspending or risking compliance.
This isn’t about picking the winner in the model wars—it’s about ensuring your clients win when all the models look the same.