GPT-5.5 is gone from ChatGPT, ChatGPT Work, and Codex as of October 14, 2026. OpenAI says the model remains reachable through the API, so existing integrations that call it directly will keep working. Inside ChatGPT, GPT-5.5 Instant became the default model on October 20, 2026. For ChatGPT Work and Codex, OpenAI lists newer models built for that work, among them GPT-6.1 Sol, which is rolling out to eligible paid plans.
The timing matters. October has turned into a crowded month for AI coding tools, with GPT-6.1 Sol already squaring off against Gemini 4 Argon on long-context tasks, and rivals shipping updates almost daily. For teams that built workflows around GPT-5.5 inside Codex, the retirement forces a decision: migrate to GPT-6.1 Sol, try the GPT-6 Sol or GPT-6 Luna models that OpenAI introduced in ChatGPT Work and Codex on October 20, 2026, lean harder on GPT-6 Astra Ultrafast where it is available, or route coding tasks to a competing model entirely. This piece breaks down what actually changed, what the retirement means for pricing and performance, and how GPT-6.1 Sol compares with the tools developers were already using before the retirement.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What OpenAI Actually Changed on October 14
The retirement notice singles out three product surfaces: ChatGPT, ChatGPT Work, and Codex. Every plan tier tied to those products loses default access to GPT-5.5 starting today. OpenAI has been explicit that this is a product-level change, not a model deprecation in the broader sense, because GPT-5.5 keeps running for developers who call it through the OpenAI API directly. That distinction is the whole story for engineering teams: anyone who wired GPT-5.5 into a custom pipeline through API keys does not need to touch their code today. Anyone relying on GPT-5.5 inside the ChatGPT interface, ChatGPT Work’s shared workspace tools, or Codex’s assisted coding environment is now defaulted onto a different model, whether they asked for it or not.
Codex is the product most directly affected for software teams. OpenAI’s 2026 developer updates already reshaped how Codex runs, letting it execute locally, remotely from a phone, or in the cloud, and today’s change strips out the GPT-5.5 option from that stack. In its place, OpenAI is pushing GPT-6.1 Sol as the default for agentic coding sessions, with GPT-6 Astra Ultrafast remaining available in the API and in ChatGPT Work and Codex specifically for Pro 500 and Enterprise plans. Smaller teams and individual developers on lower tiers do not get Astra Ultrafast access, which means GPT-6.1 Sol is effectively the only meaningfully upgraded option most Codex users will see today.
OpenAI has not published a hard shutoff date for GPT-5.5 API access, and the company’s own language leaves room for a longer tail before the model is fully retired from the API as well. That mirrors how OpenAI has handled past model transitions, where a product-level retirement arrives months ahead of full API deprecation, giving teams a buffer to migrate workloads without breaking production systems overnight. OpenAI’s October 14 model-release note says the API is unaffected, and an earlier retirement followed the same shape: GPT-5.4 and GPT-5.4 mini became unavailable in Codex for users signed in with ChatGPT on August 31, 2026, with OpenAI pointing those users to GPT-5.6 Terra and GPT-5.6 Luna, respectively, while the OpenAI API was not touched.
GPT-6.1 Sol: The Model Taking Over
GPT-6.1 Sol is pitched by OpenAI as a major upgrade over GPT-6 Sol, aimed squarely at three use cases: agentic coding, computer use, and professional work. The headline claim from OpenAI’s DevDay 2026 recap is that GPT-6.1 Sol delivers performance close to GPT-6 Astra while costing roughly one-fifth of Astra’s standard token price. For teams running high-volume coding agents, that pricing gap is the more consequential number than any single benchmark score, since agentic workflows can burn through tokens fast when a model is iterating on code, running tool calls, and checking its own output across multiple turns. GPT-6.1 Sol also ships with a default 500,000-token context window, which gives long agentic sessions considerable room before earlier work has to be dropped.
GPT-6.1 Sol Ultrafast is a faster variant of the model, and OpenAI’s Pro 500 plan at $500 per month lists both GPT-6 Astra Ultrafast and GPT-6.1 Sol Ultrafast, so teams evaluating latency-sensitive coding agents can test the fastest version of the model on that tier. The standard GPT-6.1 Sol is listed as a Work and Codex model and is rolling out to eligible paid plans.
OpenAI also used its DevDay announcements to introduce reusable development environments, which let teams start coding tasks with shared settings and permissions already configured, cutting down on repeated setup work across sessions. Separately, the Agents API now supports computer use, meaning developers can build agents that interact directly with software interfaces rather than being limited to text and code generation. Both features ship alongside the GPT-6.1 Sol rollout and are part of why OpenAI is framing this as more than a simple model swap.
Why This Is Happening Now
Model retirements inside ChatGPT and Codex are not new, but the pace has picked up through 2026 as OpenAI iterates faster on its Sol and Astra model lines. GPT-5.5’s exit follows a now-familiar pattern: a model ships, gets adopted inside OpenAI’s consumer and developer products, holds that position for a stretch of months, and then gets swapped out once a cheaper or faster successor clears internal benchmarks. What’s different this time is the specific framing around coding. OpenAI is not just replacing a general-purpose chat model, it is explicitly positioning GPT-6.1 Sol as a coding-first upgrade, which signals how central agentic coding has become to OpenAI’s product roadmap heading into 2027.
That focus lines up with a broader shift already visible across the developer tools market. Developers are now spending a majority of their working week reviewing AI-generated code rather than writing it from scratch, and separate research has found that AI coding agents are adding roughly 30% more code to projects without necessarily shipping more finished software. In that environment, a cheaper, faster coding model inside Codex is not a minor convenience, it is a direct response to how much token volume agentic coding workflows now consume on a daily basis.
GPT-6.1 Sol vs. the Rest of the Coding-Model Field
GPT-6.1 Sol does not land in an empty field. Google’s Gemini 4 Argon went globally available to developers and enterprises through the Gemini API, AI Studio, and the Gemini platform just days before GPT-5.5’s retirement, with OpenAI’s own DevDay materials explicitly benchmarking GPT-6.1 Sol against Astra-level intelligence rather than against Gemini directly. Gemini 4 Argon is aimed at long-horizon workflows including code migration, enterprise research, and cybersecurity, and carries reported introductory pricing of $2 per million input tokens and $10 per million output tokens, positioning it as a direct competitor for the same agentic coding budget that GPT-6.1 Sol is chasing.
Anthropic’s Claude Code line continues to hold a significant share of the dedicated coding-assistant market, and xAI’s Grok 4.7 has pushed its own coding benchmark scores higher at aggressive per-token pricing. The result is a market where no single lab has a clear, durable lead in agentic coding, and where pricing has become as much a competitive lever as raw benchmark performance. OpenAI’s choice to retire GPT-5.5 and push GPT-6.1 Sol’s roughly one-fifth-of-Astra pricing into the default Codex experience reads as a direct move to keep cost-conscious developers from drifting toward cheaper competitors.
Coding Model Comparison: Where GPT-6.1 Sol Fits
The table below lays out how GPT-6.1 Sol’s publicly stated positioning compares with the coding-focused models it now competes against directly inside developer workflows.
| Model | Maker | Primary Coding Use Case | Reported Pricing Position | Availability |
|---|---|---|---|---|
| GPT-6.1 Sol | OpenAI | Agentic coding, computer use, professional work | ~1/5th of GPT-6 Astra’s standard token price | Live in API; default in Codex |
| GPT-6 Astra Ultrafast | OpenAI | High-end agentic coding | Astra standard pricing tier | API, ChatGPT Work and Codex (Pro 500/Enterprise only) |
| GPT-5.5 | OpenAI | General-purpose coding (legacy) | Pre-retirement standard pricing | API only as of Oct. 14, 2026 |
| Gemini 4 Argon | Long-horizon code migration, enterprise research | $2/M input tokens, $10/M output tokens | Gemini API, AI Studio, Gemini platform | |
| Grok 4.7 | xAI | Agentic coding benchmarks | Reported at $2/M tokens | API |
Pricing details for Gemini 4 Argon and Grok 4.7 reflect figures reported at their respective launches; OpenAI has not published an exact per-token dollar figure for GPT-6.1 Sol, only the relative one-fifth-of-Astra comparison, so treat that column as directional rather than an exact quote.
What Changes for Codex Users Starting Today
For developers using Codex day to day, the practical impact depends heavily on which plan they are on. Enterprise and Pro 500 subscribers retain access to GPT-6 Astra Ultrafast alongside the new GPT-6.1 Sol default, giving them a choice between the fastest available model and the cheapest high-performing one. Everyone else loses the GPT-5.5 option inside the product, with no Astra Ultrafast fallback available on lower tiers. Plus subscribers are moved onto GPT-6.1 Sol, Pro subscribers get expanded access to it, and OpenAI’s pricing page lists it as unavailable on Free and Go, which instead get limited access in Work and Codex on desktop to GPT-5.6 Terra.
Teams running Codex in local, remote-from-phone, or cloud modes should expect the same model swap across all three deployment paths, since OpenAI has not indicated the retirement is limited to any one Codex surface. The reusable development environments feature announced alongside GPT-6.1 Sol is meant to smooth over some of this transition by letting teams preserve shared settings and permissions rather than reconfiguring from scratch every time they start a new coding task, which should reduce some of the friction that normally comes with a forced model change.
Developers who built custom tooling around OpenAI’s Responses API are among the least affected by today’s change, since that integration path calls models directly rather than through the ChatGPT or Codex product layer. The same logic applies to any pipeline built around direct API calls generally: the retirement notice is explicit that it does not extend to the API, so GPT-5.5 remains callable there for now.
Which Models Codex and ChatGPT Work Now Offer
The model lineup inside Codex and ChatGPT Work is wider than the GPT-6.1 Sol default alone. OpenAI introduced GPT-6 Sol and GPT-6 Luna in ChatGPT Work and Codex on October 20, 2026, and its release notes describe them as separate from the models available in regular ChatGPT. On October 22, 2026, both were available in the API as gpt-6-sol and gpt-6-luna. GPT-6.1 Sol is also listed as a Work-and-Codex model, and Business model controls listed GPT-6 Sol, GPT-6.1 Sol, and GPT-6 Luna for Work and Codex on October 20, 2026.
Access still depends on the plan. GPT-6.1 Sol was rolling out to eligible paid plans, and teams should check their own plan page before assuming a given model appears in their Codex picker. For anyone worried that the newer models push everything else out, OpenAI’s Help Center says the GPT-6 rollout does not retire older selectable models, and GPT-5.6 models remained selectable rather than being retired.
Pro subscribers have one more date to track. Eligible subscribers could retain the previous allowance through October 29, 2026, after which the allowance would decrease while the price remained $200 per month. Teams weighing how much Codex is worth to them can also read one customer’s account: “At Harvey, Codex transformed how we build by cutting early iteration time by 30–50%, freeing engineers to focus on system design and high-leverage decisions.”
A Pattern of Rapid Model Turnover in 2026
GPT-5.5’s retirement is not an isolated event this year. OpenAI pulled GPT-6.1 Astra over safety flaws earlier in 2026 before later reinstating an Astra-branded option in a different form, and the company has repeatedly cut API pricing on its GPT-6 line as competition from Google, Anthropic, and xAI intensified. OpenAI cut GPT-6 API prices by 50% earlier this year while simultaneously rolling out a new intelligent interface inside ChatGPT, a combination that suggested the company was already bracing for the kind of pricing pressure that Gemini 4 Argon and Grok 4.7 are now applying directly to the coding-model segment.
Historically, OpenAI has treated ChatGPT and Codex model defaults as a faster-moving layer than its API, using product-level swaps to steer the bulk of casual and default-setting users toward whichever model best serves the company’s current cost and performance targets, while leaving the API as a slower-moving, developer-controlled layer. That pattern held again today: the API is untouched, but the product-facing default moved immediately and without an opt-out for most plan tiers.
Market and Competitive Impact
The timing right next to Gemini 4 Argon’s global developer rollout is unlikely to be coincidental. Google positioned Argon for exactly the kind of long-running agentic tasks that OpenAI is now trying to lock down with GPT-6.1 Sol inside Codex, and the two companies are effectively racing to be the default coding-agent backend for the same pool of enterprise developers. Pricing is the clearest battleground: Argon’s $2/M input and $10/M output token pricing undercuts what GPT-6 Astra charged, and OpenAI’s answer is to make GPT-6.1 Sol roughly a fifth of Astra’s price rather than compete purely on capability.
For Claude Code, Codex, and GitHub Copilot, which have been splitting developer mindshare across different workflows, a cheaper default inside Codex puts pressure on competitors to match pricing rather than only compete on raw benchmark scores. Codex CLI and Claude Code have already been trading benchmark leads through 2026’s string of monthly releases, and GPT-6.1 Sol’s price-to-performance pitch gives OpenAI a new argument to make in that fight even if it does not top every coding benchmark outright.
Model Transition Pricing and Performance Snapshot
| Metric | GPT-5.5 (Retiring) | GPT-6.1 Sol (Replacement) |
|---|---|---|
| Product availability | Retired from ChatGPT, ChatGPT Work, Codex (Oct. 14, 2026) | Listed as a model for ChatGPT Work and Codex, rolling out to eligible paid plans |
| API availability | Still accessible via OpenAI API | Live in OpenAI API |
| Target use case | General-purpose assistant and coding | Agentic coding, computer use, professional work |
| Relative pricing vs. GPT-6 Astra | Not directly compared by OpenAI | ~1/5th of Astra’s standard token price |
| Ultrafast variant | Not applicable | GPT-6.1 Sol Ultrafast included in the Pro 500 plan |
What Developers Should Do Right Now
- Check which Codex plan tier your team is on, since Pro 500 and Enterprise users keep GPT-6 Astra Ultrafast access while other tiers default fully to GPT-6.1 Sol.
- Audit any workflow that assumed GPT-5.5 as the default inside ChatGPT, ChatGPT Work, or Codex, since those defaults have already changed as of today without requiring any action on your part.
- Leave direct API integrations alone for now, since OpenAI’s retirement notice explicitly excludes API-based GPT-5.5 access from today’s change.
- Benchmark GPT-6.1 Sol against your existing coding agent setup before committing fully, particularly if your workflow is latency-sensitive, since OpenAI offers GPT-6.1 Sol with Ultrafast mode on the Pro 500 plan.
- Watch whether GPT-6.1 Sol Ultrafast reaches more plans, since it is currently offered on Pro 500 and not on the Pro 100 or Pro 200 plans at launch.
Predictions: What Happens Next in the Coding-Model Race
Based on OpenAI’s release pattern through 2026 and the competitive moves already underway from Google and xAI, a few outcomes look likely over the coming months.
- GPT-6.1 Sol Ultrafast is already available. OpenAI lists it with the Pro 500 tier, and the Help Center describes an Ultrafast mode for GPT-6.1 Sol in Work and Codex.
- GPT-5.5’s API access gets a formal sunset date within the next two quarters. OpenAI’s past model transitions have typically moved from product retirement to full API deprecation within a few months once a clear successor is in place.
- Google and xAI respond with their own pricing adjustments. GPT-6.1 Sol’s one-fifth-of-Astra pricing claim puts direct pressure on Gemini 4 Argon’s per-token rates and on Grok 4.7’s positioning, and neither competitor is likely to let that pricing gap stand unanswered for long.
- Lower-tier Codex users increasingly push for Astra Ultrafast access. Restricting the fastest model to Pro 500 and Enterprise plans creates an obvious upsell path, and developer pushback over tiered access to the best-performing model is a predictable near-term friction point.
- Agentic coding benchmarks become the primary marketing battleground through early 2027. With GPT-6.1 Sol, Gemini 4 Argon, Grok 4.7, and Claude Code all competing for the same coding-agent workloads, expect each lab to publish narrower, more favorable benchmark comparisons rather than broad capability claims.
How This Fits the Bigger Picture for AI Coding Tools
Today’s retirement is a small, mechanical event on its own, a single model swapped for another inside three product surfaces. But it lands inside a much larger shift in how software gets written. Surveys already show developers spending the bulk of their working hours reviewing AI-generated output rather than authoring it directly, and separate findings suggest AI coding agents are inflating codebases with more code rather than more finished, working software. Against that backdrop, every pricing and performance change to the underlying models carries outsized weight, because it directly shapes how much AI-assisted coding costs at scale and how reliable the output is likely to be.
OpenAI’s bet with GPT-6.1 Sol is that cheaper, Astra-adjacent performance will keep developers inside its ecosystem rather than pushing them toward Gemini 4 Argon or Grok 4.7. Whether that bet pays off will show up less in benchmark charts and more in which model ends up as the quiet default behind the agentic coding workflows that are already reshaping how much of the industry’s code gets written in the first place.
Frequently Asked Questions
Is GPT-5.5 completely shut down as of October 14, 2026?
No. GPT-5.5 is retired from ChatGPT, ChatGPT Work, and Codex across all plans as of today, but it remains accessible through the OpenAI API, according to OpenAI’s own retirement notice.
What model replaces GPT-5.5 inside Codex?
GPT-6.1 Sol becomes the default model for agentic coding inside Codex. Pro 500 and Enterprise plan users also retain access to GPT-6 Astra Ultrafast as an alternative.
How does GPT-6.1 Sol’s pricing compare to GPT-6 Astra?
OpenAI says GPT-6.1 Sol delivers performance near GPT-6 Astra at roughly one-fifth of Astra’s standard input and output token pricing, though OpenAI has not published an exact dollar figure for GPT-6.1 Sol itself.
Is GPT-6.1 Sol Ultrafast available yet?
GPT-6.1 Sol Ultrafast is no longer a coming-soon feature. OpenAI’s Pro 500 plan at $500 per month now includes both GPT-6 Astra Ultrafast and GPT-6.1 Sol Ultrafast.
Does this change affect existing API integrations built on GPT-5.5?
Not directly. OpenAI’s retirement notice applies to ChatGPT, ChatGPT Work, and Codex specifically, and explicitly states the OpenAI API is not affected by today’s change.
How does GPT-6.1 Sol compare to Gemini 4 Argon for coding tasks?
Both target long-running, agentic coding workflows. Gemini 4 Argon reports introductory pricing of $2 per million input tokens and $10 per million output tokens, while OpenAI positions GPT-6.1 Sol primarily on its roughly one-fifth-of-Astra pricing claim rather than a direct published rate, making a precise dollar-for-dollar comparison difficult with currently available figures.
Will GPT-5.5 eventually disappear from the API too?
OpenAI has not announced a specific API deprecation date for GPT-5.5. Based on the company’s past model transitions, a full API sunset is plausible in future months but has not been confirmed.
Who is most affected by today’s retirement?
Developers and teams using GPT-5.5 by default inside ChatGPT, ChatGPT Work, or Codex are the most affected, since their default model has already changed. Teams with direct API integrations are largely unaffected for now.
