LLM Price War Continues: API Costs Drop 40% in Q2 2026

The LLM price war that began in late 2025 continued through Q2 2026, with average API costs dropping 40% from Q1 levels. The major providers — OpenAI, Anthropic, Google — all cut prices multiple times, and open-source alternatives continued to close the capability gap. Here's what the price cuts mean for the agent ecosystem.

The price cuts

• OpenAI: GPT-5 input costs down 35%, output costs down 40%
• Anthropic: Claude 4 Sonnet down 30%, Claude 4 Opus down 25%
• Google: Gemini 3 Pro down 45% (largest cut in the industry)
• Open source: Llama 4 inference costs on par with cheapest proprietary models

What it means for agents

Lower API costs directly improve agent economics:

• Higher ROI. Agent workflows that were marginally profitable become clearly profitable.
• More complex workflows. Multi-agent workflows that were too expensive become viable.
• Lower platform costs. Agent platforms (Lindy, Relevance) can reduce prices or improve margins.
• More experimentation. Lower costs make it easier to test new agent workflows without significant budget commitment.

Who benefits most

Custom agent builders

Companies building custom agents with frameworks like LangChain or CrewAI see direct cost reductions. A multi-agent workflow that cost $500/month to run in Q1 might cost $300/month in Q2.

Agent platforms

Platforms like Lindy and Relevance benefit from improved margins. Some of this savings may be passed to customers; we expect platform pricing to remain stable through 2026 with improved feature-to-price ratios.

Heavy users

Power users on usage-based plans (Relevance, Sierra) see the most direct benefit. Heavy users on subscription plans (Claude, OpenAI) benefit less directly — subscription prices haven't dropped, though rate limits have effectively increased as compute costs decrease.

Outlook for H2 2026

We expect the price war to continue through H2 2026, with another 20-30% in cuts likely. The open-source ecosystem (Llama, Mistral, etc.) is setting a floor on pricing that proprietary models can't ignore. By end of 2026, we expect LLM API costs to be roughly 50% below Q1 levels.

For agent users, this is unambiguously good news. Lower costs mean more experimentation, more complex workflows, and better ROI. The era of expensive AI is ending; the era of cheap AI is arriving.