OpenAI has announced an 80% price reduction for its most powerful reasoning large language model, o3, lowering costs to $2 per 1M input tokens and $8 per 1M output tokens. This price cut comes from optimizations in the inference stack without changing the model itself. OpenAI recommends using o3 for coding and tasks involving agentic tool calling, function calling, and instruction following, noting it now costs the same as GPT-4.1 and is cheaper than GPT-4o.
Alongside this, OpenAI introduced o3-pro, a more compute-intensive version designed for solving challenging problems with reliable answers. The price for o3-pro has been cut by 87%, now costing $20 per 1M input tokens and $80 per 1M output tokens. Available as ‘o3-pro-2025-06-10’ in the Responses API, o3-pro supports image inputs, function calling, and Structured Outputs. Due to its complexity, some requests may take minutes, with a new background mode available to prevent timeouts.
Tokens represent the semantic units LLMs use to process language, and pricing per million tokens is standard in API usage. This update makes o3 more accessible for developers needing advanced reasoning and positions OpenAI competitively against models like Google DeepMind’s Gemini 2.5 Pro, Anthropic’s Claude Opus 4, and DeepSeek’s reasoning suite.
