LLM Pricing Wars: DeepSeek V4, Open‑Weight Models, and the Future of Affordable AI
Review your LLM budget and evaluate open‑weight alternatives such as DeepSeek V4 or Allen AI’s OLMO for cost‑effective inference.
Review your LLM budget and evaluate open‑weight alternatives such as DeepSeek V4 or Allen AI’s OLMO for cost‑effective inference.
Summary
The article examines the steep price differences between frontier models from Anthropic and OpenAI and the low‑cost DeepSeek V4, noting a roughly 50‑fold price gap per token. It highlights that Anthropic and OpenAI have positioned their models as premium, scarcity‑driven products, while DeepSeek offers a more affordable alternative. Google released Gemma 4 in April 2026, and OpenAI’s last open‑weight GPT models appeared in 2025, whereas Anthropic has yet to release any open‑weight models. Allen AI’s OLMO model is available for download, though its data cutoff is December 2024, and the NSF/Nvidia partnership aims to develop fully open AI with open training pipelines. The discussion suggests that open‑weight models are cheaper because community testing and loss‑leader pricing reduce costs. It also raises concerns about potential regulatory restrictions on open‑weight models due to geopolitical fears. The article calls for a shift toward truly open AI that includes open data pipelines.
Key changes
- Google Gemma 4 released April 2026.
- DeepSeek V4 offers low‑cost inference, ~50× cheaper per token than Anthropic/OpenAI frontier models.
- OpenAI last released open‑weight GPT models in 2025; Anthropic has no open‑weight releases.
- Allen AI’s OLMO model is available for download, though data cutoff is Dec 2024.
- NSF/Nvidia partnership aims to develop fully open AI with open training pipelines.
- Open‑weight models are cheaper due to community hardware testing and loss‑leader pricing.
- Pricing strategies of Anthropic/OpenAI create scarcity and premium branding.