AI news · Tuesday, September 22, 2026
Anthropic and OpenAI Trade Blows With New, Cheaper AI Model Tiers
The AI labs are effectively engaged in a price war, with both OpenAI and Anthropic launching new, more efficient model tiers this week. OpenAI introduced updated versions of its 'Sol' and 'Luna' models, claiming they offer GPT-6 Astra-level factuality at significantly lower costs. OpenAI (claimed) that these models outperform Anthropic's top-tier Claude models on various benchmarks while hitting price points that are 50% lower than their previous generation.
Not to be outdone, Anthropic released Claude Opus 5. 5, which the company says matches the performance of its own high-end Fable 5. 1 model while costing 40% less to run.
Anthropic is clearly trying to shed the 'Claudish' reputation of its previous models, promising that Opus 5. 5 uses less jargon and communicates more directly. Both companies are facing intense pressure to reduce costs as enterprise adoption stalls and Chinese rivals offer dirt-cheap alternatives that are rapidly climbing the performance rankings.
Xiaomi, for instance, just released its MiMo-V2. 6 model, which tops independent benchmarks for open models while costing a fraction of what Western labs charge, though Anthropic has accused Xiaomi of illegally distilling its proprietary data to get there. Meanwhile, the hardware layer of the AI boom is getting crowded and commoditized.
Qualcomm just launched new smartphone chips—the Snapdragon 8 Elite Gen 6 and an 'Extreme' version—designed to run AI agents locally rather than in the cloud, signaling that chipmakers are betting on phones becoming the primary interface for personal AI. This shift toward local processing is already changing the workforce, as companies like Salesforce and Wipro push tools that turn employees into managers of AI agents. Workers are increasingly finding themselves in a weird middle ground where they spend 30% or more of their day supervising 'agents'—bot fleets that need constant hand-holding—without the commensurate pay raise or management title that usually accompanies such oversight.
The tension is also hitting traditional sectors like data center construction, where local residents are pushing back against the 'inevitable' expansion of infrastructure. Researchers found that public opposition to data centers is growing organic and post-partisan, fueled by concerns over electricity bills and the secretive playbook developers use to push projects through town councils. Even in space, the industry is pivoting toward autonomy to solve the problem of expensive ground control, with startups like AstroForge now putting AI in command of spacecraft to avoid the $200 million price tag of building their own antenna network.
The quick hits
- OpenAI and Anthropic both launched new, cheaper AI model tiers this week — the race to the bottom on price is officially on for enterprise and developer tools.
- Qualcomm is pushing AI onto your phone with its new 2nm chips — the goal is to run personal agents locally instead of relying on expensive cloud servers.
- Data center expansion is hitting a wall as local resistance grows — residents are increasingly organized and skeptical of the industry's 'inevitable' utility-scale projects.
- Employees are becoming 'agent managers' by default — workers are now spending hours supervising AI bots but aren't getting the pay or titles that come with management roles.