Jiberish

The Bench

OpenAI, ChatGPT, and dots: models, plans, and our verdict

OpenAI's ChatGPT, GPT-6 models, Codex, and dots: where they stand, plan and API prices, data terms, and dated recent changes.

Kyle Del FranciaUpdated 17 min read
In this piece

OpenAI on the bench

  • AdoptChatGPT & GPT-6: ChatGPT for people and teams, Codex for code, and the GPT-6 family in the API.Strongest at: Everyday knowledge work · Broad workforce rollout · Cost-tiered models
  • Trialdots: Always-on agents in ChatGPT, each with its own cloud computer and browser, reachable in Slack and Teams.Strongest at: Background tasks · Ongoing monitoring · Enterprise residency options
See every verdict on the bench

OpenAI makes ChatGPT, the GPT-6 family of models, the Codex coding agent, and dots, the always-on agents in ChatGPT. It sells to individuals, businesses, and developers through a wide set of surfaces, each with its own models, limits, and data terms. This page covers where OpenAI stands, the models and products that exist today, what they cost, how data is handled, and what changed recently. We checked every source linked here on November 10, 2026, mostly OpenAI's developer documentation, its ChatGPT Learn pages, and its trust portal. A few figures appear only on OpenAI's marketing site or Help Center, which we could not retrieve; where that leaves a gap, we say where to look. ChatGPT features vary by plan, workspace settings, and rollout, and API documentation does not establish what a ChatGPT account includes.

Where it stands

ChatGPT and the GPT-6 models are on Adopt, and dots are on Trial. These are our judgments from hands-on use and from OpenAI's own documentation.

ChatGPT earns Adopt as a general-purpose assistant for people and teams. It covers writing, analysis, and multi-step tasks through ChatGPT Work, which uses a member's authorized files, apps, and tools (Work overview). Projects give a set of chats the same uploaded files, instructions, and connected sources (projects), and Business workspaces carry the commitment that matters most for company data: "No training on your business data by default" (pricing). The GPT-6 family behind it runs from Astra, OpenAI's most capable model, to Luna, its most efficient, with 6.1 Sol and 6.1 Terra between them and a 1,050,000-token context window across all four in the API (models). Codex brings the same models to software work in the desktop app, web, CLI, and IDE.

Dots stay on Trial. A dot keeps working between conversations on its own cloud computer and browser, and OpenAI now describes a "family of frontier models that power dots," recently expanded to include GPT-6.1 Terra (dots). The documented gaps have narrowed since our last check. Dots are generally available in eligible Enterprise, Edu, and Healthcare workspaces, members there get access by default from November 18 unless the workspace opts out, and billing for enabled dots begins the same day (admin guide). OpenAI's dots security guide now lists data storage residency in several regions and inference residency in the US and Europe, with the UAE excluded, and says HIPAA is "supported for eligible workspaces" (dots security). What keeps them on Trial: Pro and Business Premium access is still rolling out, the security guide still labels dots Beta, the admin guide describes flat-rate or consumption billing for Enterprise without publishing prices, and neither dots nor Work provides strict zero data retention. We would move them to Adopt once billing is live with published terms, the Beta label is gone, and our own pilots hold up on recurring work.

OpenAI's main trade-off is pace. Developer tools retire on short notice, with v1/prompts, Agent Builder, and the Evals dashboard and API all shutting down on November 30, and GPT Image 2 now scheduled to follow on February 10, 2027 (deprecations). The right model depends on the surface, the plan, and the rollout, so teams need to check what their own accounts include.

The lineup today

OpenAI's model catalog lists four flagship GPT-6 models. Each has a 1,050,000-token context window and 128,000 maximum output tokens.

ModelAPI identifierOpenAI's descriptionAPI release
GPT-6 Astragpt-6-astra"Our most capable model for the most demanding work"September 3
GPT-6.1 Solgpt-6.1-sol"Near-Astra performance for complex work at a lower cost"September 29
GPT-6.1 Terragpt-6.1-terra"A fast, efficient model for visual reasoning, precise computer use, and clear, natural writing"October 26
GPT-6 Lunagpt-6-luna"Our most efficient model for focused, high-volume tasks"September 22

Release dates come from the API changelog. The GPT-6 guide notes that GPT-6 supports Structured Outputs, and that of the GPT-6 models only 6.1 Terra, GPT-6 Sol, and Luna accept a reasoning effort of none; Astra and 6.1 Sol do not. The earlier gpt-6-sol and the GPT-5.6 family remain on the pricing page. The o-series is being retired: o1, o1-pro, o3-mini, and o4-mini shut down on October 23, 2026, and o3, o3-pro, and gpt-5 are scheduled to shut down on December 11 (deprecations).

The catalog also lists image, realtime, and transcription models, plus specialized models behind trusted-access programs. Two of those arrived in the past two weeks. GPT-6 Rosalind (gpt-6-rosalind), for life sciences research, is limited to organizations approved for OpenAI's trusted-access program (changelog, October 30). GPT-6-Cyber (gpt-6-cyber), for investigating and reproducing vulnerabilities in authorized test environments and verifying fixes, requires Daybreak Red access and approval and works only on Chat Completions and Responses (GPT-6-Cyber model page, released November 5). For images, GPT-Image-2.6 Sunburst ("for detailed visual work") and Flare ("for fast iteration") launched on November 4 with SVG and animated GIF generation (models, changelog).

SurfaceWhat it isModels, per OpenAI's docs
ChatGPT chatThe chat app on personal and workspace plansChat models are listed in OpenAI's Help Center; the pricing page marks 6.1 Sol and 6.1 Terra as not in Chat
ChatGPT WorkMulti-step tasks using a member's files, apps, and tools (overview)The four GPT-6 models, varying by plan, client, and rollout (models)
CodexOpenAI's coding agent in the desktop app, web, CLI, and IDEThe same four; Free and Go can choose GPT-6 Luna in the desktop app when available (models)
dotsAn always-on agent in ChatGPTA family of frontier models, now including GPT-6.1 Terra; a dot can draw on different models for different parts of its work (dots)
API and Agents APIPer-token model access; the Agents API runs OpenAI-managed, long-running agents on the Codex harness (guide)API: the full catalog. Agents API: public beta since September 10

In Enterprise and Edu workspaces, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna stay off until an admin enables them (workspace model availability).

Several developer tools end soon. The v1/prompts endpoint and reusable prompt objects shut down on November 30, 2026, as do Agent Builder (ChatKit remains) and the Evals dashboard and API, whose existing evals went read-only on October 31 (deprecations). Further out, tts-1, tts-1-hd, and the gpt-4o-mini-tts snapshots shut down on January 6, 2027; older gpt-4o, gpt-4o-mini, and gpt-4.1 snapshots on January 14; gpt-5.5-cyber-preview on February 1; gpt-image-2 on February 10, with the GPT-Image-2.6 models as replacements; and gpt-5.3-codex, gpt-5.4-nano, and gpt-5.1 on April 1 (same page). In ChatGPT, GPT-5.5 retired on October 14 from ChatGPT, Work, and Codex (ChatGPT updates).

Plans and pricing

Prices below are list prices in US dollars as of November 10, 2026, from OpenAI's ChatGPT Learn pricing page, which says ChatGPT Work and Codex are included in ChatGPT plans. Dots eligibility comes from the dots page.

PlanList priceNotes from OpenAI's docs
Free$0Codex for quick coding tasks; GPT-6 Luna in the desktop app when available
Go$8/monthCodex for lightweight coding tasks
Plus$20/monthCodex on web, CLI, IDE extension, and iOS; GPT-6.1 Sol and GPT-6 Luna
Pro$100, $200, or $500/monthNo five-hour limit currently; Ultrafast on Pro $500; dots rolling out to users 18 and older outside the EEA, UK, and Switzerland
Business$20/user/month billed annually, $25 billed monthly2+ users; "No training on your business data by default"
Business PremiumNot on the pages we could openDots rolling out in supported ChatGPT regions, subject to workspace eligibility
Enterprise and EduCustom, through salesSCIM, EKM, RBAC, Compliance API, retention and residency controls; Edu is credit-based (education plans); dots generally available in eligible workspaces, billed from November 18
ChatGPT for HealthcareNot on the pages we could openNamed on OpenAI's HIPAA page (see below); dots generally available in eligible workspaces

For Business Premium and Healthcare prices, check OpenAI's ChatGPT pricing page or its sales team. For Plus and Standard Business, the pricing page estimates local messages per five-hour period: GPT-6 Astra 5–45, GPT-6.1 Sol 15–160, GPT-6 Sol 15–150, GPT-6.1 Terra 35–285, and GPT-6 Luna 350–3,000, and says weekly limits may also apply. On Pro and Business Premium, conversations with a dot don't count toward ChatGPT usage limits, but tasks it starts in Work or Codex do (dots). For Enterprise, the dots admin guide says eligible workspaces can pay a flat rate based on the type of work a dot does or pay by consumption, that the preview is free until billing begins on November 18, and that tasks a dot delegates to Work or Codex are charged separately at standard rates; it gives no prices (admin guide). Since October 28, eligible Pro and Business subscribers can claim up to $500 a month in API credits by linking an API organization they own (changelog).

API prices per million tokens, Standard tier, as of November 10, 2026 (API pricing):

ModelInputCached inputOutputBatch or Flex, input / output
gpt-6-astra$10.00$1.00$50.00$5.00 / $25.00
gpt-6.1-sol$2.00$0.10$10.00$1.00 / $5.00
gpt-6.1-terra$1.00$0.05$5.00$0.50 / $2.50
gpt-6-luna$0.10$0.01$0.50$0.05 / $0.25

Requests over 272K input tokens cost more (Astra: $20.00 input, $75.00 output). Fast mode, formerly Priority processing, costs $20.00 / $100.00 on Astra; Ultrafast is listed only for Astra ($60.00 / $300.00) and 6.1 Sol ($12.00 / $60.00). Regional processing endpoints, and FedRAMP endpoints, add 10% for models released on or after March 5, 2026. Cache writes are billed separately: the pricing table lists 30-minute writes at 1.25 times the input price and 12-hour writes, available in the Responses API since October 20, at 2.5 times (changelog). The trusted-access models cost more: gpt-6-cyber is $20.00 input and $100.00 output, and gpt-6-rosalind $15.00 and $75.00, for requests up to 272K input tokens. Both GPT-Image-2.6 models are priced at $8.00 input and $30.00 output per million image tokens. The API pricing page says GPT-5.6 Sol's promotional pricing is available until November 30, 2026, while the ChatGPT pricing page still says "at least through November 21."

Data, privacy, and compliance

On the API, OpenAI's data guide says data sent to the API "is not used to train or improve OpenAI models" unless you opt in, and abuse-monitoring logs are kept for up to 30 days. Zero Data Retention (ZDR) and Modified Abuse Monitoring need OpenAI's prior approval through its sales team. ZDR forces store to false on Responses and Chat Completions, but its scope has exclusions: the Agents API, Files, and Vector Stores are not ZDR-eligible and keep application state until deleted, background mode stores responses for roughly 10 minutes, and data sent to an MCP server follows that server's retention policy.

In ChatGPT, the pricing page states no training on Business data by default, and the dots security guide says "Enterprise data is not used to train models by default" (dots security). Personal-plan training settings are documented in OpenAI's Help Center, which we could not retrieve; we recommend handling customer data in a Business or Enterprise workspace. In Work, deleting a chat "schedules permanent deletion within 30 days," subject to published exceptions (Work overview), and the Compliance Logs Platform retains data for 30 days (Work admin FAQ). Deleting a dot deletes its own chats, subagent history, and scheduled tasks within 30 days (dots security), but files, Codex threads, and conversations it created are stored separately and must be removed where they live (admin guide). Neither dots nor Work with local computer access provides strict zero data retention, and dots are unavailable in FedRAMP workspaces (dots security).

For residency, the API offers regional storage in ten regions and regional processing in six: the US, Europe, the UAE, Australia, Japan, and India. Australia, Japan, and India process only gpt-6-luna and require active ZDR, the UAE needs additional approval, and any region other than the US requires approval for abuse-monitoring controls (data guide). In ChatGPT, Enterprise and Edu include residency controls. For dots, the security guide lists data storage residency in Australia, Canada, Europe, India, Japan, and the US, and inference residency in the US and Europe (dots security); UAE is not included, and workspaces with UAE data or inference residency can't use dots (dots). Work is not supported with UAE inference residency.

For HIPAA, eligible API organizations can accept OpenAI's Business Associate Agreement in Platform settings since October 5 (changelog), though the data guide says web search with live internet access is not covered by a BAA. On the ChatGPT side, OpenAI's HIPAA configuration page names ChatGPT for Healthcare, ChatGPT for Clinicians, and Regulated workspaces with a BAA. It says the BAA doesn't cover Codex cloud and that third-party plugins need their own BAA. The dots security guide now says HIPAA is supported for eligible workspaces and tells customers to confirm that their BAA and configuration cover the intended workflow (dots security). No page we opened documents BAA coverage for Work, so we would keep patient information out of it until OpenAI does.

OpenAI's trust portal states ISO/IEC 27001:2022 conformance for OpenAI's API, ChatGPT Enterprise, and ChatGPT Edu services, with ISO/IEC 27017, 27018, and 27701 control sets. Its latest SOC 2 Type 2 report covers July 1, 2025 to June 30, 2026, with products in scope on a separate status page. The portal also lists SOC 3, ISO/IEC 42001:2023, CSA STAR, FedRAMP 20x, TX-RAMP, GDPR, CCPA, and PCI DSS v4.0.1 for ChatGPT's delegated-payment components. The ISO scope does not name Business or personal plans.

What changed recently

All dates are 2026, from the API changelog, deprecations page, ChatGPT updates, and Codex changelog.

  • August 26: The Assistants API shut down.
  • Week of August 31: GPT-6 Astra arrived in Codex and ChatGPT Work; its API release followed on September 3.
  • September 10: The Agents API entered public beta.
  • September 22: GPT-6 Sol and GPT-6 Luna launched in the API and began rolling out in Codex.
  • September 29: GPT-6.1 Sol launched, and the Agents API gained computer use. OpenAI's updates for the week of September 28, alongside an undated DevDay 2026 section, include dots, ChatGPT Space, Team Tasks, and Codex Security Cloud as a research preview.
  • October 5: API organizations could accept a BAA in product.
  • October 14: GPT-5.5 retired from ChatGPT, Work, and Codex. OpenAI also gave notice that older gpt-4o, gpt-4o-mini, and gpt-4.1 snapshots shut down on January 14, 2027.
  • October 20: 12-hour prompt caching arrived in the Responses API.
  • October 22: The Decisions API became generally available with gpt-6-luna.
  • October 23: o1, o1-pro, o3-mini, and o4-mini shut down, and the Agents API gained scheduled automations.
  • October 26: GPT-6.1 Terra launched in the API, and in Work and Codex on Plus, Pro, Business, Enterprise, and Edu.
  • Week of October 26: Texting a dot by iMessage or RCS entered beta for Pro users in the US.
  • October 29: Codex CLI 0.166.0 added GPT-6.1 Terra to its bundled, Vertex AI, and Amazon Bedrock model catalogs.
  • October 30: GPT-6 Rosalind launched for approved life sciences organizations. The Agents API gained tool updates in existing sessions, background compaction by default, and automatic review of tool and permission requests.
  • October 31: Existing evals became read-only ahead of the November 30 shutdown.
  • November 2: Outbound SIP calling through the Live API opened to all organizations.
  • Week of November 2: Dots on iOS gained replies from a notification and sharing photos from other apps.
  • November 4: GPT-Image-2.6 Sunburst and Flare launched in the API, with SVG and animated GIF generation.
  • November 5: GPT-6-Cyber launched for organizations with Daybreak access, and OpenAI scheduled gpt-5.5-cyber-preview to shut down on February 1, 2027.
  • November 6: Codex CLI 0.167.0 added an agent command center and backend switching from the session picker.
  • November 10: OpenAI scheduled gpt-image-2 to shut down on February 10, 2027.

Where it fits

ChatGPT fits organizations that want a familiar assistant for a broad workforce: Business for smaller teams, and Enterprise or Edu where SCIM, enterprise key management, role-based access, the Compliance API, and retention and residency controls matter (pricing). Teams already on ChatGPT can extend into Codex for software work without adding a vendor.

On the API, the four GPT-6 models make it practical to match cost to difficulty: Luna for focused, high-volume tasks, 6.1 Terra for visual reasoning, computer use, and natural writing at speed, 6.1 Sol for complex work at a lower cost than Astra, and Astra for the hardest problems (models). Structured Outputs and the Agents API, in public beta, support products that need predictable data or long-running agents (GPT-6 guide, Agents API). Security and life sciences teams with trusted access now have dedicated models in GPT-6-Cyber and GPT-6 Rosalind.

Dots fit recurring work that benefits from an agent that keeps going, such as watching an inbox, a Slack channel, or a set of sources, once they reach your plan. Enterprise, Edu, and Healthcare admins should decide before November 18 whether to opt out or set roles, since access turns on by default and billing starts that day (admin guide). Healthcare organizations should read OpenAI's HIPAA configuration closely: coverage is tied to named workspaces, eligible configurations, and API terms, not to every surface (HIPAA configuration).

What we're watching

Dots are the call most likely to change. Since November 1, OpenAI has documented dots residency, HIPAA support for eligible workspaces, and Microsoft Teams messaging (dots), and the Enterprise admin guide now calls access a preview rather than a beta setting. We would consider moving them to Adopt once billing starts on November 18 with terms we can verify, the security guide drops its Beta label, Pro and Business Premium rollouts finish, and our own pilots hold up on recurring work. Still unreleased or limited: texting a dot is a US-only Pro beta, Free and Go web Plugin Extensions are "coming soon," commercial Sign in with ChatGPT is in a limited trial with selected partners, and Codex Security Cloud is a research preview (ChatGPT updates).

On the developer side, we are tracking the November 30 shutdowns of v1/prompts, Agent Builder, and Evals; the end of GPT-5.6 Sol's promotional pricing, which the API pricing page dates to November 30; the December 11 retirement of o3, o3-pro, and gpt-5; the January and February 2027 retirements of older gpt-4o, TTS, cyber-preview, and gpt-image-2 models; and whether the Agents API leaves beta and becomes ZDR-eligible. We will add Business Premium pricing and Enterprise dots prices once they appear on a page we can verify.

From verdict to deployment

Putting OpenAI to work is the hard part.

We design and build AI into the websites and applications a business runs on, automate workflows with agents that work across its tools, and roll AI out across whole organizations, with every team trained on its real work. We build on whichever lab fits the job, and we stay through launch and adoption.

Book a free call

Put this into practice

Bring the decision you’re working through.

A free discovery session is a chance to review your goal, the tools or website you have, and the help you need. We’ll identify a sensible next step and discuss scope before any project commitment.

Custom AI Systems & Agents →AI Advisory & Training →Book a free call