o3 — reviews, specs & pricing

OpenAI's advanced reasoning model for complex multi-step problems.

Summary

o3 is a reasoning-focused model that uses extended chain-of-thought to solve complex math, science and coding problems. It supports tool use during reasoning. It's aimed at the most demanding analytical tasks.

Sample use case

Used for advanced research, complex coding tasks, and multi-step agentic reasoning. Suited for scientific and mathematical problem solving.

Specifications

  • Provider: openai
  • License: closed
  • Context: 200k tokens
  • Input price: $2/M tok
  • Output price: $8/M tok
  • Released: 2025-04-16

Pros

  • State-of-the-art reasoning
  • Tool use during thought
  • Strong benchmarks

Cons

  • Higher latency
  • Expensive at scale
  • Closed weights

Average rating 3.8 from 5 community reviews on Reviuws.

Benchmark results

BenchmarkScoreLatency
LMArena (Chatbot Arena) Elo1408.0%
AIME 202491.6%
GPQA Diamond87.7%
SWE-bench Verified69.1%
Humanity's Last Exam20.3%

Measured by the Reviuws Bench suite on identical prompts. Compare these scores against every other model.

Community reviews

Ars Technica on o3

Rating: 4.0 / 5 — by Benj Edwards (use case: Complex agentic reasoning and research tasks)

Edwards covers o3 and o4-mini as a meaningful step forward in giving reasoning models full access to ChatGPT's tools, framing it as OpenAI's most capable reasoning release to date with genuine, though incremental, utility gains.

Pros: First reasoning models with full tool access (browsing, code execution, image reasoning); strong benchmark performance

Cons: Complexity and cost of tool-integrated reasoning; benefits vary by task type

TechCrunch on o3

Rating: 4.0 / 5 — by Maxwell Zeff (use case: Technical problem-solving in math, coding, and science)

Zeff reports o3 as OpenAI's most advanced reasoning model yet, outperforming prior models on math, coding, and science benchmarks, while noting the broader competitive pressure from rival reasoning models.

Pros: State-of-the-art reasoning benchmark scores; pauses to work through problems methodically

Cons: High cost relative to non-reasoning models; competitive landscape closing the gap quickly

Simon Willison's blog on o3

Rating: 4.0 / 5 — by Simon Willison (use case: Agentic multi-tool reasoning workflows)

Willison highlights o3's emphasis on agentic tool use as its standout new capability, noting it can autonomously decide when and how to invoke web search, code execution, and image analysis mid-reasoning.

Pros: Sophisticated autonomous tool orchestration; strong multimodal reasoning integration

Cons: Behavior and cost trade-offs still being worked out relative to sibling models

TechRadar on o3

Rating: 3.0 / 5 — by Eric Hal Schwartz (use case: Logical reasoning and puzzle-solving comparisons)

Schwartz's head-to-head logic tests found o3's results inconsistent relative to GPT-4.1 and GPT-4o, concluding that raw model comparisons don't always align with expectations set by benchmarks.

Pros: Sometimes produced sharper, more rigorous logical reasoning than peers

Cons: Inconsistent performance across similar logic puzzles; results didn't always match hype

Simon Willison's blog on o3

Rating: 4.0 / 5 — by Simon Willison (use case: Abstract reasoning benchmark evaluation)

Willison highlights Francois Chollet's early-access analysis of o3's breakthrough ARC-AGI score as the most substantive coverage available, underscoring both the achievement and its very high compute cost.

Pros: Breakthrough score on ARC-AGI reasoning benchmark; genuinely novel reasoning capability

Cons: Extremely expensive to run in high-compute mode ($6,677 for the public eval set)

Frequently asked questions

What is o3?

o3 is a reasoning-focused model that uses extended chain-of-thought to solve complex math, science and coding problems. It supports tool use during reasoning. It's aimed at the most demanding analytical tasks.

How much does o3 cost?

o3 costs $2 per million input tokens and $8 per million output tokens.

Is o3 open source?

o3 is released under the closed license.