OpenAI: GPT, Codex, and the Most Widely Used AI Products

Learning Objectives

• Identify OpenAI's key models and their respective strengths and ideal use cases
• Explain OpenAI's structure as a mission-driven company with commercial pressure and its implications
• Compare OpenAI's models against each other to guide selection decisions

OpenAI: From Research Lab to AI Giant

OpenAI was founded in December 2015 as a nonprofit AI research organization, backed by Elon Musk, Sam Altman, Peter Thiel, and others including Reid Hoffman and the founders of PayPal. The founding mission was explicit: develop safe artificial general intelligence for the benefit of humanity, and counter the concentration of AI capabilities in a small number of corporate hands.

The irony that OpenAI is now one of those concentrated hands is not lost on critics.

In 2019, OpenAI created a "capped-profit" subsidiary to attract commercial investment, with Microsoft becoming the primary investor — ultimately committing over $13 billion. Sam Altman serves as CEO. The company released GPT-2, GPT-3, GPT-4, and then ChatGPT in November 2022 — which became the fastest-growing consumer application in history, reaching 100 million users in two months.

ChatGPT triggered the current wave of AI investment and product development across the entire industry. Nearly every AI product you will encounter has some ancestral connection to the cultural and commercial moment that ChatGPT created.

Corporate Restructure and Funding (2026)

OpenAI's corporate structure has undergone a major transformation. The for-profit subsidiary has been restructured as OpenAI Group PBC — a public benefit corporation — while the nonprofit was renamed the OpenAI Foundation, which holds equity currently valued at approximately $130 billion, making it one of the best-resourced philanthropic organizations in history.

In February 2026, OpenAI closed a $110 billion funding round at a $730 billion pre-money valuation — the largest private funding round in history. Key investors: SoftBank ($30 billion), NVIDIA ($30 billion), and Amazon ($50 billion committed). Total funding now exceeds $168 billion, with annual recurring revenue growing from roughly $2 billion in January 2024 to approximately $20 billion.

OpenAI is actively laying groundwork for a 2026 IPO, with internal targets of an H2 2026 filing and a 2027 listing.

The April 2026 Microsoft Amendment

On April 27, 2026, OpenAI and Microsoft announced a major restructuring of their seven-year partnership — reshaping what is arguably the most consequential commercial relationship in AI:

• Cloud exclusivity ends. Microsoft's exclusive intellectual-property license on OpenAI's models becomes non-exclusive through 2032. OpenAI can now ship products on any cloud — AWS, Google Cloud, Oracle, or its own infrastructure. Microsoft remains the primary partner with first-shipping rights when it has the capacity to support new models.
• Revenue-share reversed. Microsoft stops paying revenue share to OpenAI. OpenAI continues paying Microsoft a capped revenue share through 2030.
• Equity preserved. Microsoft retains approximately 27% equity in the new OpenAI Group PBC — valued at roughly $135 billion post-October 2025 recapitalization — keeping it one of the largest single shareholders.
• $250 billion Azure commitment. OpenAI contractually commits to spending $250 billion on Microsoft Azure services, ensuring Microsoft remains the dominant infrastructure partner even as exclusivity formally ends.

The amendment effectively transitions OpenAI from a Microsoft-bound subsidiary-style relationship to a more independent — but still deeply intertwined — strategic partner. The practical effect: ChatGPT and the OpenAI API can now run on multiple clouds (relevant for compliance, redundancy, and customer-of-choice deployments), while the $250 billion Azure commitment guarantees the partnership remains commercially central for years.

The GPT-5 Family: OpenAI's Flagship Models

GPT-5.5 — The New Agentic Flagship (April 2026)

GPT-5.5 is OpenAI's latest flagship model, released April 23, 2026 — just six weeks after GPT-5.4. OpenAI describes it as "a new class of intelligence built specifically for real work and for powering agents." It replaces GPT-5.4 as the default model in ChatGPT for Plus, Pro, Business, and Enterprise users. The OpenAI API rollout follows in the coming weeks with separate safeguards for autonomous use.

Key capabilities and benchmarks (vs. GPT-5.4):

• Terminal-Bench 2.0: 82.7% (GPT-5.4: 75.1%) — multi-tool command-line workflows requiring planning and error recovery; the metric coding-agent builders watch most closely
• GDPval: 84.9% (GPT-5.4: 83.0%) — economically valuable knowledge-work tasks measured against expert humans
• OSWorld-Verified: 78.7% (GPT-5.4: 75.0%) — desktop UI automation
• FrontierMath Tier 4: 35.4% (Claude Opus 4.7: 22.9%, Gemini 3.1 Pro: 16.7%) — the hardest open math benchmark, where GPT-5.5 currently leads frontier models by a wide margin
• Token-efficient: uses significantly fewer tokens than GPT-5.4 on the same Codex tasks at matched latency, lowering long-agent-run cost
• ~1 million token context (922K effective) carried over from GPT-5.4; Codex variant uses 400K tuned for long coding sessions
• GPT-5.5 Pro: 90.1% on BrowseComp — highest score for an autonomous web-research agent

Variants:

• GPT-5.5 — default for Plus, Pro, Business, and Enterprise
• GPT-5.5 Thinking — extended reasoning mode; available on Plus and above
• GPT-5.5 Pro — highest reasoning effort; Pro, Business, and Enterprise only

OpenAI did not ship mini or nano GPT-5.5 variants at launch. Older 5.x models (5.1, 5.2, 5.3, plus the existing GPT-5.4 mini) remain available for cost-sensitive serving.

Pricing: ChatGPT Plus ($20/month) is the entry point for GPT-5.5; Pro ($200/month) unlocks GPT-5.5 Pro. API endpoints are reported at $5 input / $30 output per million tokens for gpt-5.5 and $30 / $180 for gpt-5.5-pro — roughly 2x the per-token cost of gpt-5.4. Pricing pending official OpenAI publication when API endpoints go live.

When to use GPT-5.5: Multi-step agentic workflows, knowledge-work automation, scientific and mathematical reasoning, long-running coding agents — anywhere GPT-5.4 was the default. Note that Claude Opus 4.7 still leads SWE-bench Pro (64.3% vs. 58.6%) for the hardest end-to-end software-engineering tickets.

GPT-5.5

OpenAI

Closed

Strengths

New agentic flagship; Terminal-Bench 2.0 82.7%; GDPval 84.9%; FrontierMath Tier 4 35.4%; token-efficient agents; Thinking + Pro variants

Context Window

~1 million tokens (API)

Pricing

ChatGPT Plus $20/mo; Pro $200/mo; API ~$5/$30 per million tokens (rolling out)

OpenAI →

GPT-5.4 — Prior Flagship (Lineage)

GPT-5.4 was OpenAI's flagship from March 5 to April 23, 2026. It introduced the 1 million token context window, native computer-use, the Thinking / Pro / mini / nano variant lineup, and a 33% reduction in false claims vs. GPT-5.2. It remains available as gpt-5.4 via the API and continues to power the Free and Go tier mini model. GPT-5.1 models were deprecated on March 11, 2026.

The Codex Family: OpenAI's Coding Models

GPT-5.3-Codex — The Most Powerful Coding Model

GPT-5.3-Codex is OpenAI's most capable coding model, designed specifically for agentic software development tasks. It powers the OpenAI Codex platform — a web-based coding agent that can implement features, debug code, write tests, and create pull requests from natural language descriptions.

Key characteristics:

• 25% faster than GPT-5.4 at coding-specific tasks
• Full agentic coding capability: takes a task description and autonomously completes multi-step implementation
• Deep codebase understanding: can reason across large codebases, not just individual files
• Best-in-class for: implementing new features, refactoring code, and systematic debugging

When to use: Complex coding projects, multi-file implementation, when you want the maximum coding capability OpenAI offers.

GPT-5.3-Codex

OpenAI

Closed

Strengths

Most powerful OpenAI coding model; 25% faster than GPT-5.4 at coding tasks; full agentic multi-step implementation

Context Window

200K tokens

Pricing

Available via OpenAI Codex platform

OpenAI →

GPT-5.3-Codex-Spark — Real-Time Coding

GPT-5.3-Codex-Spark is a specialized model optimized for one property above all others: speed.

Running on Cerebras hardware (which uses a different chip architecture than NVIDIA for extreme inference speed), it delivers:

• 1,000+ tokens per second — approximately 10x faster than typical hosted model inference
• Sub-second response time for most coding queries
• 128K context window
• The "first real-time coding model" — fast enough to feel like native IDE autocomplete

When to use: IDE integration where latency matters, any workflow where you are iterating rapidly and waiting for responses is a bottleneck, high-volume applications where cost-per-query is important.

GPT-5.3-Codex-Spark

OpenAI

Closed

Strengths

1,000+ tokens/sec on Cerebras; sub-second response; real-time coding experience; 128K context

Context Window

128K tokens

Pricing

Available via API

OpenAI →

GPT-OSS: OpenAI Goes Open Source

GPT-OSS marks a significant strategic shift for OpenAI: the release of a competitive open-weight model.

Released under Apache 2.0 license (permissive, including commercial use), GPT-OSS:

• 20 billion+ parameters — significant capability in a relatively efficient size
• Designed for on-premise deployment and edge computing
• Available for download and self-hosting
• Apache 2.0 license allows commercial deployment without royalties

This release was widely interpreted as a response to Meta's Llama and DeepSeek's open-source momentum — OpenAI demonstrating that it can play in the open-source space while maintaining its closed flagship models.

When to use: Organizations that need full data control, on-premise deployment, fine-tuning for specific domains, or edge deployment where cloud API access is unavailable or undesirable.

GPT-OSS

OpenAI

Open Source

Strengths

Apache 2.0 license; 20 billion+ params; on-premise and edge deployment; commercial use allowed

Context Window

128K tokens

Pricing

Free (open weights); compute costs for self-hosting

OpenAI →

Creative Models: GPT Image 2

GPT Image 2 — Flagship Image Generation (April 2026)

GPT Image 2 (also branded ChatGPT Images 2.0 inside ChatGPT) is OpenAI's current flagship image generation model, released on April 21, 2026. It replaced GPT Image 1.5 as the default image model and is available through ChatGPT and Codex immediately; the gpt-image-2 API opens to developers in early May, with a parallel rollout in Microsoft Foundry on Azure.

Key characteristics:

• First reasoning-native image model: GPT Image 2 brings OpenAI's O-series "thinking" into image generation — the model plans composition, typography, and spatial relationships before drawing, significantly raising success rates on complex scenes
• Multilingual text rendering: Character-level accuracy for Japanese, Korean, Chinese, Hindi, and Bengali inside generated images — a major unlock for non-Latin-script content
• Web-search integration: Can search the web during the planning phase to incorporate real-time facts (new products, recent events) that postdate the model's training cutoff
• 2K resolution + 8 images per prompt: Higher-quality finals and faster A/B exploration than prior versions
• Integrated into ChatGPT: available directly in conversation without switching tools
• Image Arena #1 by +242 points: within 12 hours of launch, the largest recorded lead on that leaderboard

GPT Image 1.5 remains the prior flagship; DALL-E 3 was the generation before that.

Sora — Text-to-Video (Discontinued March 2026)

Sora was OpenAI's video generation model, capable of producing cinematic-quality video with synchronized audio from text prompts. First demonstrated in February 2024 and released publicly in December 2024, Sora was integrated into ChatGPT and available as a standalone app.

On March 24, 2026, OpenAI announced Sora's discontinuation — shutting down the iOS app, API, and sora.com. The company cited the need to prioritize compute resources for enterprise products and research. The Disney partnership (which included plans for a $1 billion investment and character licensing) collapsed alongside the shutdown. Sora's research team continues at OpenAI, redirected toward world simulation for robotics.

OpenAI's Ecosystem Position

Beyond its individual models, OpenAI's ecosystem position is significant:

• ChatGPT is the world's most used AI consumer product — enormous brand awareness and user familiarity
• OpenAI API is the default integration for many developers — extensive documentation, SDKs, and community
• Microsoft partnership means OpenAI models are available through Azure OpenAI Service, reaching enterprise customers via Microsoft's existing relationships
• Plugin and tool ecosystem: OpenAI's function calling and tool use API is widely implemented across enterprise software

The OpenAI developer ecosystem is the largest in the AI industry. This matters for practical decisions: more tutorials, more Stack Overflow answers, more open-source examples, more developer knowledge on the market.

Talent Acquisitions

OpenAI's strategy extends beyond model development to acquiring key talent from the open-source AI ecosystem. In February 2026, Peter Steinberger — creator of OpenClaw, the fastest-growing open-source project on GitHub with over 247,000 stars — joined OpenAI. Sam Altman described Steinberger as someone who would "drive the next generation of personal agents." OpenClaw is a free, open-source autonomous AI agent that runs locally and uses messaging platforms (WhatsApp, Telegram, Discord, Slack) as its interface. The project moved to an independent open-source foundation after the hire, but the acquisition signals OpenAI's intent to build personal agent capabilities beyond coding (Codex) and research (ChatGPT).

Key Takeaways

• OpenAI offers a comprehensive portfolio spanning general intelligence (GPT-5.5), agentic coding (GPT-5.3-Codex, Codex-Spark), open weights (GPT-OSS), and image generation (GPT Image 2 — the first reasoning-native image model, April 2026); video generation (Sora) was discontinued in March 2026
• GPT-5.5 (released April 23, 2026) is the new agentic flagship: leads frontier models on Terminal-Bench 2.0 (82.7%), GDPval (84.9%), OSWorld-Verified (78.7%), and FrontierMath Tier 4 (35.4%); three variants (standard, Thinking, Pro)
• GPT-5.4 (March 2026) remains available via the gpt-5.4 API endpoint and powers the Free and Go tier mini model — it kept its position as flagship for only six weeks before GPT-5.5 launched
• GPT-5.3-Codex for maximum coding capability; Codex-Spark when speed is critical
• GPT-OSS is OpenAI's first open-weight model (Apache 2.0) — a strategic response to Meta Llama and DeepSeek
• ChatGPT's dominance as a consumer product and the OpenAI API's developer ecosystem are competitive advantages beyond raw model capability
• OpenAI restructured as a public benefit corporation (OpenAI Group PBC) in 2026, raising $110 billion at a $730 billion valuation — with an IPO expected in 2027