OpenAI Sets January Deadline: GPT-4 API Retires, GPT-6 Luna Is the Replacement

30-Second Summary

OpenAI has given enterprise operators a 90-day deadline. Certain GPT-4 series models will be removed from the OpenAI API on January 14, 2027. Any application, product, or workflow that calls a deprecated model after that date will receive errors instead of responses. OpenAI's recommended migration path is GPT-6 Luna, released September 22, 2026, priced at $0.10 per million input tokens and $0.50 per million output tokens.

At a Glance

• Topic: Enterprise AI infrastructure
• Company: OpenAI
• Date: October 14, 2026
• Announcement: Certain GPT-4 series models removed from OpenAI API on January 14, 2027
• What Changed: OpenAI formally communicated the sunset date via its release notes, giving operators 90 days to migrate
• Why It Matters: Any GPT-4 API integration left unmigrated will break on the deadline date, with no fallback
• Who Should Care: Any organisation running applications, automations, or internal tools that call the OpenAI API directly

Key Facts

• Deprecation date: January 14, 2027
• Announcement date: October 14, 2026
• Recommended replacement: gpt-6-luna
• GPT-6 Luna input pricing: $0.10 per million tokens (Source: OpenAI, September 2026)
• GPT-6 Luna output pricing: $0.50 per million tokens (Source: OpenAI, September 2026)
• GPT-6 Luna release date: September 22, 2026, alongside GPT-6 Sol
• Performance comparison: GPT-6 Luna at high effort improved by 5.4 percentage points over GPT-5.6 Luna on AutomationBench (Source: OpenAI, vendor-reported, API-environment results)
• Enterprise activation: Administrators on OpenAI Enterprise plans must enable GPT-6 models in Settings before members can access them

What Happened

OpenAI updated its release notes on October 14, 2026, with a formal notice that certain GPT-4 series models will be removed from its API on January 14, 2027. The company identified GPT-6 Luna as the recommended replacement. The announcement is consistent with OpenAI's broader model lifecycle policy, which periodically retires older models as newer ones reach equivalent or superior capability at lower cost.

GPT-6 Luna launched on September 22, 2026, alongside GPT-6 Sol. The two models represent OpenAI's GPT-6 mid-tier and were released to ChatGPT Plus, Pro, Business, Enterprise, and Edu subscribers on a rolling basis. Luna is positioned as the cost-efficient option within the GPT-6 family, while Sol targets higher-complexity tasks. Both are available via the API for developers and enterprise teams.

The January 14 date gives operators exactly 90 days from the announcement to migrate. That deadline includes the end-of-year holiday period, which compresses the effective working window for many organisations. Enterprise teams with formal software development lifecycles, change management requirements, or vendor approval processes will need to begin migration immediately to avoid a forced last-minute change.

The deprecation affects organisations that have integrated GPT-4 models directly into their own applications, not just teams accessing ChatGPT via a browser. Internal tools, customer-facing AI features, automated workflows, and data pipelines that call the OpenAI API by model name are all affected. Teams using GPT-4 through platforms such as Microsoft Copilot, which manage the underlying model independently, may not need to take action, but should confirm this with their platform vendor.

Why It Matters

The API is where enterprise integration lives. Most meaningful enterprise AI work happens via the API, not in a chat interface. The deprecation affects companies that have embedded AI into their own products, not consumer users of ChatGPT. If your organisation built internal tools, customer-facing automation, or data workflows on GPT-4 models, those systems will stop working on January 14, 2027.

90 days is less time than it sounds. The announcement came on October 14. The deadline is January 14. That window spans November planning cycles, December holidays, and January quarter-start. For organisations that require testing, change approval, and staged rollouts, the effective window is closer to six to eight weeks. Teams that wait until December will be scrambling.

The cost case for migrating is strong. GPT-6 Luna is priced at $0.10 per million input tokens, which is dramatically lower than GPT-4 pricing from prior generations. Operators who move their GPT-4 API usage to GPT-6 Luna will typically see a significant reduction in their monthly AI infrastructure spend at the same or greater capability level.

Performance has not been traded away. GPT-6 Luna records a 5.4 percentage point improvement over its predecessor on OpenAI's AutomationBench evaluation. These are vendor-reported results from API-environment testing, but they indicate that Luna is not a capability downgrade from GPT-4. For most business writing, summarisation, data extraction, and classification tasks, Luna performs at or above GPT-4 standards.

Enterprise admins have a specific task. OpenAI Enterprise plans require administrators to actively enable GPT-6 models in Settings before users can access them. This is not automatic. An organisation that delays the administrator step will find that GPT-6 Luna is unavailable to its team even after migrating code to call the new model name.

The migration signals the pace of the AI stack refresh cycle. GPT-4 launched in March 2023. It is being retired in January 2027. That is roughly a four-year cycle from release to retirement, but the practical active life for enterprise integration was shorter. Operators should assume that enterprise AI models will turn over on roughly 18 to 24-month cycles going forward, and build their integration architecture to accommodate planned migration.

The David and Goliath View

OpenAI is accelerating the pace at which enterprise teams must stay current with their AI stack. The GPT-4 retirement is the clearest signal yet that building on a specific model version is a technical debt risk, not just a capability limitation. Operators who have treated AI as a one-time implementation are about to learn the same lesson the cloud migration taught the previous decade: the platform evolves, and standing still means breaking.

For most operators of teams between 10 and 200 people, this is also the nudge they needed to run a proper AI audit. Many organisations have GPT-4 integrations in production that were set up by a contractor or a single team member, documented partially, and never reviewed. The January deadline forces that review to happen, which is ultimately a healthy exercise. A migration plan for one model becomes the template for managing the next one.

The cost reduction in moving to GPT-6 Luna is real and should be communicated upward in any organisation where AI spend is a budget line. This is not a routine upgrade that costs more, it is a required migration that saves money. That framing matters for getting resource allocation and approval quickly, which is what operators need right now given the deadline.

Where This Fits in the AI Stack

• API layer: Any application calling gpt-4, gpt-4-turbo, or related model names directly must update to gpt-6-luna before January 14, 2027
• Prompt engineering: GPT-6 Luna accepts the same message format as GPT-4. Most prompts migrate without changes. Prompts that rely on specific GPT-4 behaviours should be tested
• Embeddings: The deprecation applies to chat completion and other GPT-4 endpoints. Embedding model retirements are announced separately
• Cost management: At $0.10 per million input tokens, GPT-6 Luna enables higher-volume use cases that were cost-prohibitive on GPT-4
• Platform integrations: Operators using AI through Salesforce, Microsoft 365, Google Workspace, or other managed platforms should check with those vendors about whether they handle model updates automatically

Questions Operators Are Asking

Do we have to do anything if we only use ChatGPT through a browser? No. The deprecation affects the OpenAI API, which developers use to build integrations. If your team only accesses AI through ChatGPT's chat interface, OpenAI manages the underlying model and no action is required.

Will our applications break on January 14 if we do not migrate? Yes. API calls to deprecated model names will return errors after the retirement date. Any application that calls a deprecated model by name will stop working until the code is updated.

Is GPT-6 Luna as capable as GPT-4 for our use case? For most business tasks, including writing, summarisation, data extraction, classification, and customer communication, GPT-6 Luna performs at or above GPT-4 standards. The safest approach is to run both models on your existing prompt library and compare outputs before switching production traffic.

How long will GPT-6 Luna be supported? OpenAI has not published a support timeline for GPT-6 models. Based on the GPT-4 lifecycle, operators should plan for a roughly three to four year window, but the pace of model development has accelerated and future cycles may be shorter.

What if we cannot complete the migration before January 14? Contact OpenAI enterprise support as early as possible. OpenAI has historically provided short extensions in cases where enterprise customers have documented timelines and active migration work underway. Do not wait until December to make that contact.

Citable Summary

OpenAI announced on October 14, 2026 that certain GPT-4 series models will be removed from its API on January 14, 2027. The recommended replacement is GPT-6 Luna, available at $0.10 per million input tokens and $0.50 per million output tokens. Operators with GPT-4 API integrations have 90 days to migrate. Enterprise plan administrators must separately enable GPT-6 models in Settings before their teams can access them. GPT-6 Luna recorded a 5.4 percentage point improvement over its predecessor on OpenAI's AutomationBench evaluation in vendor-reported API-environment testing.