← Back to hard AIs

Verify critical details — pricing, licensing, availability — with the model's source before business decisions. Full methodology →

Models

OpenAI

4.5 ★ — The broadest model lineup in the industry — a frontier closed flagship, a deep pricing ladder, and (since gpt-oss) a credible open-weight line under Apache 2.0. Held off a 5 by premium flagship pricing and a still-mostly-closed catalog, but few creators give buyers this many options.

Type
ai-native-company
Country
US
Founded
2015
License posture
mixed
Website
https://openai.com

Quick Take

The company that kicked off the modern AI boom with ChatGPT, OpenAI fields the widest lineup anywhere: a closed frontier line (GPT and o-series) plus, since 2025, an open Apache 2.0 family (gpt-oss).

Who They Are

OpenAI is the AI-native company whose ChatGPT launch in late 2022 turned large language models into a mainstream tool, and it remains the most widely used AI provider in the world. Its business is built around the closed, hosted frontier: you rent its models through the OpenAI API or use them inside ChatGPT, rather than downloading them. Microsoft is a major partner and investor, and OpenAI's models are also available through Microsoft's Azure OpenAI Service.

For business readers, the defining feature is breadth. OpenAI offers more models at more price points than anyone — from the flagship GPT-5.5 down through mid-tier and ultra-cheap "mini" and "nano" variants, plus the specialized o-series reasoning models and Codex coding models. The single biggest lever on your OpenAI bill is simply picking the right model for the task rather than defaulting to the flagship.

Model Philosophy

OpenAI's center of gravity is closed and hosted — the GPT and o-series models are proprietary, API-only, and priced as a premium product. But in August 2025 it broke a long open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. drought (its first since the GPT-2 era) by releasing gpt-oss-120b and gpt-oss-20b under the permissive Apache 2.0 license. That gave OpenAI a genuine two-track posture: a closed frontier for the highest-stakes work, and open weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself. you can self-host for privacy, control, and cost.

The closed line emphasizes agentic execution and developer tooling — function calling, structured outputs, long context, and coding agents. The open line is explicitly aimed at self-hosting and on-deviceRunning a model directly on a consumer device — a laptop, a phone, a smart speaker — rather than in a data center. On-device inference keeps data private by never leaving the device, and works offline. Small models (under ~10B parameters, often quantized) can run on-device; larger models cannot yet. use, with the trade-off that open weights carry a different safety profile (once released, they can't be revoked or further mitigated).

What To Know Before You Commit

Decide whether you're renting or owning, then pick the right rung on the ladder. If you want frontier capability through an API, OpenAI's depth is the draw — but note the flagship is among the pricier options (GPT-5.5 sits above Claude Opus on output cost, and DeepSeek undercuts everyone at the top tier), so route cheaper tasks to mini/nano models. If you want to self-host, the gpt-oss models are the relevant line, and their Apache 2.0 license is as clean as open licensing gets.

On data governance, OpenAI is US-based with enterprise and zero-retention options for API traffic and Azure hosting for regulated buyers; the open gpt-oss weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself. sidestep routing entirely. The main caution is cost discipline: the o-series bills internal reasoning tokens at output rates, which can multiply costs, and the newest flagships use breakpoint pricing that rises above ~272K tokens.

How They Compare

Against Google and Anthropic, the other closed Western frontier labs, OpenAI's edge is lineup breadth and ecosystem reach (ChatGPT's user base, Azure, a huge developer tool ecosystem); Google counters with native multimodality and longer context, Anthropic with coding and safety positioning. Against the open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. players — Meta, Qwen, DeepSeek, and Google's Gemma — OpenAI's gpt-oss now competes directly on Apache-licensed open weightsThe numerical values inside a trained model that encode everything it has learned. A model is, functionally, a giant list of weights — tens of billions of numbers for a mid-sized model, hundreds of billions for a frontier model. "Open-weight" means those numbers are published. "Downloading the weights" means getting the actual file you'd need to run the model yourself., though those rivals often field larger or more varied open models. The honest summary: nobody offers more total options than OpenAI, but it's rarely the cheapest at any given tier.

Original Models

Gpt 5 5

OpenAI's frontier flagship: top-tier coding, reasoning, and vision with a 1.05-million-tokenThe basic unit of text a model reads and writes. Tokens are roughly three-quarters of a word in English — so 100 tokens is about 75 words. Models don't see letters or words directly; they see tokens. Pricing is almost always quoted per million tokens, and context windows are measured in tokens rather than words. context windowThe maximum amount of text the model can "see" at once — prompt plus prior conversation plus any documents you give it. Measured in tokens (which are roughly three-quarters of a word each). A 128K context window is about 96,000 words of input — roughly a 400-page book. Larger context windows let the model work with bigger documents but cost more to run. — the best of the GPT line, at premium prices.

OpenAI's most capable — and by far most expensive — model: research-grade reasoning for the narrow band of problems where getting the answer right outweighs the cost.

Gpt 5 4

OpenAI's value frontier: the previous flagship, still the right default for most production work — near-GPT-5.5 capability at half the price.

GPT-5.4 Mini — a cheap, fast closed tier (~$0.75/$4.50) for routed production traffic.

GPT-5.4 Nano — OpenAI's ultra-budget tier (~$0.20/$1.25) for high-volume simple tasks.

Gpt 5 2

GPT-5.2-Codex — a dedicated coding-agent model (~$1.75/$14) for code-generation API flows.

Gpt 5

GPT-5 — the prior-generation flagship, multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. and capable, now behind the 5.4/5.5 line.

GPT-5 Mini — the cheaper multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. GPT-5 tier, superseded by the GPT-5.4 mini line.

Gpt Oss

OpenAI's open comeback: an Apache 2.0 reasoning modelA model trained to "think through" problems step by step before answering, often by producing internal reasoning that's either shown or hidden from the user. Reasoning models trade speed for accuracy on hard problems — they're slower and more expensive per answer, but markedly better at math, logic, and complex analysis. OpenAI's o1 series and Mistral's Magistral are reasoning models. that nears o4-mini quality, runs on a single 80GB GPUThe specialized chip that runs most AI models. Originally designed for 3D graphics, GPUs turned out to be excellent at the math AI requires. Nvidia dominates the AI GPU market; common datacenter models include the H100, H200, and B200. Running an AI model without a GPU is possible but painfully slow for anything but the smallest models., and you can download, fine-tuneA model that has been further trained on additional data to specialize it for a particular task, domain, or style. Fine-tuning a general model on medical literature produces a medical specialist; fine-tuning on your company's support tickets produces a support assistant that sounds like your team. Fine-tunes are much cheaper to create than training a model from scratch., and self-host freely.

OpenAI's on-deviceRunning a model directly on a consumer device — a laptop, a phone, a smart speaker — rather than in a data center. On-device inference keeps data private by never leaving the device, and works offline. Small models (under ~10B parameters, often quantized) can run on-device; larger models cannot yet. open model: o3-mini-class reasoning that runs locally on 16GB of memory, under clean Apache 2.0 — download it and run it on a good laptop.

O Series

o3 — an o-series step-by-step reasoning modelA model trained to "think through" problems step by step before answering, often by producing internal reasoning that's either shown or hidden from the user. Reasoning models trade speed for accuracy on hard problems — they're slower and more expensive per answer, but markedly better at math, logic, and complex analysis. OpenAI's o1 series and Mistral's Magistral are reasoning models.; strong on hard problems, now alongside newer reasoning tiers.

o4-mini — a cheaper o-series reasoning modelA model trained to "think through" problems step by step before answering, often by producing internal reasoning that's either shown or hidden from the user. Reasoning models trade speed for accuracy on hard problems — they're slower and more expensive per answer, but markedly better at math, logic, and complex analysis. OpenAI's o1 series and Mistral's Magistral are reasoning models. (~$1.10 input); the reference point for gpt-oss-120b.

Gpt 4 1

GPT-4.1 — an older 1M-context multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. generalist, still available for compatibility.

GPT-4.1 Nano — an ultra-cheap text tier (~$0.10/$0.40) for simple high-volume tasks.

Gpt 4o

GPT-4o — OpenAI's 2024 multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. workhorse (text/image/audio), now legacy but still widely used.

Sources