one year on
OpenAI launches o3-mini reasoning model for free ChatGPT users
The model arrives days after DeepSeek’s R1 stunned the industry, positioning itself as a cheaper, faster alternative that also marks the first reasoning model available to free-tier ChatGPT users.
OpenAI today released o3-mini, a new reasoning model that for the first time gives free ChatGPT users access to a model that fact-checks itself before answering. The launch, originally previewed in December alongside the more capable o3, comes days after Chinese AI lab DeepSeek shook the industry with its R1 reasoning model, which OpenAI has accused of being built on stolen IP.
O3-mini is tuned for STEM tasks — programming, math, and science — and OpenAI claims it matches or outperforms its o1 family while running 24% faster and costing 63% less than o1-mini. In the API, o3-mini is priced at $0.55 per million cached input tokens and $4.40 per million output tokens, undercutting o1-mini but still above DeepSeek R1’s $0.14 and $2.19. Users can choose low, medium, or high reasoning effort depending on their need for speed versus accuracy; free ChatGPT users get a “Reason” button built into the chat bar.
On benchmarks, o3-mini only beats R1 on AIME 2024 and SWE-bench Verified when using high reasoning effort, and still trails R1 on GPQA Diamond at low effort. The model does not yet support image analysis. OpenAI positions o3-mini as a specialized alternative to the broader o1, writing that it “provides a specialized alternative for technical domains requiring precision and speed.”
The release is widely read as OpenAI’s direct response to DeepSeek’s disruptive debut — one that forced the company to move fast on both pricing and accessibility, even as it prepared for what could be one of the largest funding rounds in history. The chat forums tonight are lit up with comparisons between o3-mini-high and R1. For now, the narrative is that OpenAI is playing catch-up on cost, not intelligence.
Reported that OpenAI is battling perception it is ceding ground to Chinese companies like DeepSeek, which OpenAI alleges might have stolen its IP. The launch comes as OpenAI tries to shore up its Washington relationship and pursue a massive data center project.
view the original post →One year later — open only if you can handle spoilers
o3-mini helped reset the pricing conversation around reasoning models, but DeepSeek R1 continued to claim a vocal open-weights community. OpenAI later integrated o3-mini's reasoning-effort slider into its broader product, though the model never quite displaced the buzz around R1's transparency.