The Great AI Perception Divide
The Great AI Perception Divide. Chinese Open Source Closes the Gap with GLM 5.1.
The Great AI Perception Divide
@karpathy dropped a crucial observation about the growing gap in how people perceive AI capabilities [5][6]. Users stuck on free or outdated models are still dealing with hallucinations and basic errors, while professionals using paid tools like Codex 5.3 and Claude Code are seeing genuinely useful technical work.
He attributes this rapid progress to reinforcement learning with verifiable rewards and the industry's shift toward B2B focus [5]. When you can actually verify if code works or if an agent completed a task correctly, the feedback loops get much tighter.
This explains why AI discourse feels so polarized. Half the conversation is people frustrated with ChatGPT 3.5 from 2023, while the other half is building production systems with models that actually work. The gap is only widening.
Chinese Open Source Closes the Gap with GLM 5.1
Zhipu AI just released GLM 5.1 under MIT license, and it's topping open source leaderboards while matching closed models like Claude Opus 4.6 and GPT-5.4 [7][8][9]. The model excels at coding and agentic tasks, particularly long-horizon work, at roughly 6x cheaper pricing than the big names.
@bindureddy called it the new open source leader and specifically recommended it for coding agents [10][11]. This isn't just another "good for open source" release—it's genuinely competitive with the best proprietary models.
The implications are massive. If open source can deliver GPT-5.4 level performance at a fraction of the cost, the entire pricing structure of AI services is about to get very interesting. The moat around frontier models is shrinking fast.
What This Means For Your Business
The AI landscape is stratifying rapidly. Free and cheap tiers are becoming genuinely inadequate for serious work, while the premium tools are reaching genuine utility. If you're still evaluating AI based on free ChatGPT, you're making decisions with outdated information. The models that matter for business applications—coding, agentic tasks, complex reasoning—live behind paywalls now.
The emergence of capable open source alternatives like GLM 5.1 creates a third path: deploy your own infrastructure and avoid the subscription treadmill entirely. For companies building AI-powered products, this could mean the difference between sustainable unit economics and bleeding money on API calls.
The shift from general-purpose models to specialized tools is accelerating. Rather than asking "which AI is best," start asking "which model is best for coding, which for voice, which for video." The age of one-size-fits-all AI is ending, and the winners will be those who orchestrate the right tools for the right tasks. Key takeaway: Stop optimizing for the cheapest AI and start optimizing for the most effective AI for each specific use case.
Sources
- https://www.cnbc.com/2026/04/09/openai-chatgpt-pro-subscription-anthropic-claude-code.html
- https://techcrunch.com/2026/04/09/chatgpt-pro-plan-100-month-codex
- https://x.com/sama/status/2042342572958630332
- https://x.com/bindureddy/status/2042093196948467929
- https://x.com/karpathy/status/2042334451611693415
- https://www.threads.com/@carnage4life/post/DW7erkSj944/andrej-karpathy-nails-the-perception-gap-in-ai-capabilities-that-ive-mentioned
- https://huggingface.co/zai-org/GLM-5.1
- https://llm-stats.com/models/glm-5.1
- https://artificialanalysis.ai/models/glm-5-1
- https://x.com/bindureddy/status/2041566630212403476
- https://www.reddit.com/r/ClaudeCode/comments/1rf3obx/new_banger_from_andrej_karpathy_about_how_rapidly
Stay ahead of AI
No spam. Unsubscribe anytime.
Want to go deeper?
Reading the news is one thing. Exploring the frontier is another. See what we're building.