o3 Mini High

o3 Mini High is OpenAI's reasoning-focused variant in the o family, exposed through OpenRouter under a distinct model key. It is functionally the same underlying model as the base o3-mini, distinguished only by having its reasoning effort parameter set to high, which allocates more compute to chain-of-thought inference. The variant is positioned as a cost-efficient option for STEM-heavy workloads, with particular strengths in science, mathematics, and coding, and exposes three adjustable reasoning effort levels so teams can trade latency and cost against depth of deliberation.

Training reportedly combines supervised chain-of-thought fine-tuning with reinforcement learning from verifiable rewards, an approach aimed at producing more reliable multi-step reasoning while keeping the model compact enough for cost-sensitive deployment. Developer-facing capabilities include function calling, structured outputs, and streaming, making it well suited to agentic and tool-augmented pipelines that need tight response formatting. Practical fit centers on workloads that benefit from deliberate, high-effort reasoning over a long context rather than broad multimodal coverage, with the variant delivering deeper deliberation on math, code, and analytical reasoning tasks when that extra inference budget is warranted.

OpenRouteropenai/o3-mini-higho

Open provider page Provider docs

Quick Info

Provider: OpenRouter
Model key: openai/o3-mini-high
Release date: Feb 12, 2025
Last updated: Feb 12, 2025
Knowledge cutoff: 2023-10-31
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.10
Output token cost: $4.40

Limits

Output tokens: 100,000 tokens
Context window: 200,000 tokens

Transparent token rates

Compare o3 Mini High pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about o3 Mini High

OpenRouter

CoverageBenchmark

A peer-reviewed study in Journal Français d'Ophtalmologie (Volume 49, Issue 1, January 2026) compared ChatGPT o3-mini-high against DeepSeek-R1 on 55 text-based ophthalmology case-solving questions spanning nine subspecialties, using Elstein's multi-component diagnostic reasoning framework and a five-point Global Qualit ChatGPT o3-mini-high correctly answered 80% of the questions versus 54.5% for DeepSeek-R1 (P < 0.001), with a Cohen's kappa of 0.462 for inter-model agreement, and required additional prompts far less often (2 vs. 12 cases). Both models performed best on retina/vitreous cases and worst on glaucoma cases, indicating dom

OpenRouter

CoverageBenchmark

A Scientific Reports study analyzes the relationship between reasoning chain length and accuracy across o1-mini and o3-mini variants on the Omni-MATH benchmark. The authors find that o3-mini (medium) achieves higher accuracy than o1-mini without requiring longer reasoning chains, and that accuracy generally declines as For the high-reasoning variant, the paper observes that o3-mini (high) gains only a marginal accuracy improvement over o3-mini (medium), but does so by allocating substantially more reasoning tokens across all problems—including those the medium variant already solves. The findings suggest newer o-series reasoning mode

OpenRouter

CoverageBenchmark

According to Magica's model profile, o3 Mini High is developed by OpenAI with a 200K-token input context window and a 100K-token output limit, and was released on February 12, 2025, with a stated knowledge cutoff of October 31, 2023. The model is listed at $1.10 per million input tokens and $4.40 per million output tok The profile reports capability flags including function calling, structured output, reasoning mode, content moderation, and an always-enabled reasoning mode defaulting to effort "High." On Artificial Analysis-style indices it scores 11 on Intelligence, 16.3 on Coding, and 0.9 on Agentic; these figures are third-party a