DeepSeek’s low-cost dominance triggers mass switches from pricey US models and even prompts Zhipu to hike GLM prices right into irrelevance.
Users worldwide are ditching OpenAI and Anthropic’s Claude in droves for DeepSeek, the Chinese upstart that redefined AI economics with dirt-cheap tokens and open-source power. Reports from app stores show ChatGPT uninstalls spiking amid backlash to OpenAI deals, while Claude tops download charts as a stopgap, only for DeepSeek to pull ahead on price and performance. Now Zhipu AI’s GLM models face the same fate at home, with recent price jumps handing DeepSeek even more ammunition.
The shift started when DeepSeek revealed advanced models at fractions of US costs, shrinking the gap to months instead of years. As Bloomberg noted in their coverage of OpenAI accusations, Chinese firms like DeepSeek aggressively distill rivals’ outputs to match capabilities without the R&D burn. Developers and startups, squeezed by GPT-4o or Claude Opus tabs, find DeepSeek’s V3.2 at $0.25 per million input tokens impossible to ignore.
Zhipu tried undercutting DeepSeek earlier with GLM-4.5 at 11 cents per million inputs, but flipped to hikes of 8-17% on GLM-5.1 just last month, chasing monetization amid losses over 1.7 billion yuan last year. Reuters detailed the coding plan jump of 30% in February, citing demand, yet timing couldn’t be worse with DeepSeek’s V4 preview crushing code benchmarks. Users see the move as desperate, accelerating flight to platforms that prioritize affordability over quarterly profits.
At home, this pits Chinese giants against each other in a brutal cost war. Economic Times highlighted GLM-4.5’s agentic edge on lighter hardware, but DeepSeek’s transparency and lower barriers win everyday tasks from startups to enterprises. Zhipu’s raises mirror Alibaba and Tencent adjustments, yet they risk ceding ground as DeepSeek’s open-source ethos draws global devs.
Global Ripple Hits US Leaders
Even Claude and OpenAI feel the heat. Anthropic blocked DeepSeek after spotting 24,000 fake accounts distilling Claude data, per VentureBeat, while OpenAI cries foul on model theft. But users care less about origins than outputs, with DeepSeek handling reasoning and code at scales US firms charge premiums for. The New York Times pegged Chinese open-source models at a big chunk of 2025 apps, a trend exploding into 2026.
For AI builders, DeepSeek proves low-cost innovation levels the field. No longer must teams burn cash on proprietary APIs when open alternatives match 90% of needs. Watch Zhipu for response, perhaps a GLM price rollback or new efficiency plays, but momentum favors disruptors who keep tokens cheap. The real win lies in what comes next: agentic breakthroughs anyone can afford, reshaping who leads AI deployment worldwide.
Also read: The Microsoft-OpenAI breakup is playing out in slow motion and both sides are accelerating it • OpenAI is building an AI smartphone and Qualcomm just got the most important chip contract in years • Discord sleuths breach Anthropic’s zero-day hunter in a wake-up call for contained AI security