OpenAI shipped Workspace Agents inside ChatGPT Business, Enterprise, Edu, and Teachers earlier this year. The free promotional window closed on May 6, 2026. The credit-based pricing model went live the same day.
As of Berlin morning, May 11, 2026, there is still no public per-credit dollar amount.
OpenAI has confirmed publicly that credits are consumed by agent complexity, tools called, and execution time. It has confirmed that credits also unlock Deep Research, Thinking models, Image Gen, Advanced Voice, and Codex. It has not published the credit-to-dollar rate, the per-action cost, the monthly credit pool per seat, or the overage rate.
That is the entire public rate card after five full days of paid availability. OpenAI's typical Help Center publication lag for new commercial features is twenty-four to seventy-two hours. We are now at roughly three times that window.
If you run procurement for a B2B SaaS company or you sit on the GEO team that recommended this rollout to operations, this silence is your problem, not OpenAI's. Here is how to think about it.
What OpenAI actually shipped
The product is straightforward to describe. The pricing is not.
Workspace Agents are persistent ChatGPT agents that take on multi-step research and operational tasks across a workspace's connected apps. They run on the same plan tiers most teams already pay for: ChatGPT Business at twenty-five dollars per user per month, Enterprise at roughly sixty dollars per user per month, plus Edu and Teachers. The free window covered the rollout phase and ended May 6.
After May 6, every agent run consumes credits. Credits scale with three variables.
- •Agent complexity. A heavier reasoning chain costs more than a simple summary.
- •Tools called. Web search, file retrieval, code execution, and connector reads all cost more than a single internal step.
- •Execution time. Longer runs cost more than shorter runs.
That is the public model. The number that converts those three variables into a euro or dollar amount has not been published.
We covered the surface implications of Workspace Agents in ChatGPT Workspace Agents are running research on your category right now and in your prospect's sales team has a ChatGPT agent that researches every lead. This post is about a different question: what does the pricing silence mean for the buyer.
Why the pricing silence is unusual
OpenAI publishes commercial rate cards quickly, even for opaque-by-design products. Codex has a public rate card. API pricing is on the pricing page. The Workspace Agents gap is the longest enterprise rate-card silence OpenAI has run in the last twelve months that we have tracked.
OpenAI publishes commercial rate cards inside seventy-two hours. Workspace Agents is at five days and counting. That is signal, not delay.
There are five reasons that gap matters.
Reason #1: Five days is three times OpenAI's normal envelope
The publication-norm baseline is twenty-four to seventy-two hours. Day five is past that. The expected behavior would be a Help Center article within seventy-two hours, even if the article is technical and qualified. The absence of any document at all is the data point.
Reason #2: Credit math is the only way buyers can model spend
A flat per-seat number is forecastable. A credit-based model with no public conversion rate is not. Procurement teams cannot build a budget line for "agent complexity, tools called, and execution time" without a unit cost. They can sign for ChatGPT Business at twenty-five dollars per seat. They cannot sign for credits priced at some-number-OpenAI-will-tell-you-later.
Reason #3: The three credit variables compound
A research agent that calls four tools across a forty-minute run draws from all three credit variables at once. The cost of a "typical" workflow depends on the team. A finance analyst running spreadsheet reconciliation will burn very different credit volumes from a marketing analyst writing a brief. Without a rate card, every team estimates differently, and procurement loses any ability to standardize.
Reason #4: Competitors all published rate cards on launch day
Anthropic published Claude API pricing on launch. Google publishes Gemini Enterprise per-seat numbers. Microsoft publishes Copilot per-seat add-on pricing on the marketing page. The pattern across the rest of the category is to publish at launch. OpenAI is the outlier.
Reason #5: Usage is accruing while the rate card is missing
Credits started consuming on May 6. Every workspace running Workspace Agents since then has accumulated usage that will be invoiced against a rate that has not been disclosed. This is functionally different from a free trial. It is paid consumption at an unpublished price.
What this means for B2B SaaS procurement
The risk is not the absolute dollar amount. Workspace Agents will eventually price somewhere sane. The risk is the absence of forecastable unit cost during the window when teams are building usage patterns.
Three procurement risks stack here.
Risk #1: Hidden budget overshoot for early-adopter teams
Teams that piloted Workspace Agents in March or April had every reason to assume the eventual rate would be reasonable. Those teams have now spent six weeks training employees to use agents heavily. Behavior is set. If the May 6 rate card lands at a price that makes high-frequency use expensive, the team is stuck between cutting back on a tool people already rely on and absorbing a budget line that was not in the plan.
Risk #2: Vendor-lock asymmetry
Six weeks of agent usage builds workspace dependencies. Saved agents. Connector configurations. Internal prompts. Standardized handoffs between human and agent steps. The longer the rate card stays opaque, the more sunk cost the buyer accumulates before they see a price.
Risk #3: Procurement-policy gaps
Most B2B SaaS procurement policies require a published rate card before sign-off for any new commercial line item. The Workspace Agents product is live, in production, and consuming dollars, with no document procurement can attach to the requisition. That is a policy violation for many companies, not a minor process gap.
Is your team building dependencies on a product without a rate card?
We audit B2B SaaS AI-surface adoption against procurement standards, then map the visibility and spend risk across ChatGPT, Claude, Gemini, Perplexity, and Copilot. Most clients see surface-specific spend and visibility data inside thirty days.
Get Your AI Visibility AuditHow GEO and AI visibility teams should respond
The procurement angle is the obvious one. The GEO angle is the one most teams will miss. Workspace Agents are a citation surface, beyond their role as an internal productivity tool. Your brand may be in those agents' research output right now, and your visibility there is independent of the pricing question.
Here are five steps to take this week.
Step 1: Track Workspace-Agents-cited brands in your category
A Workspace Agent running competitive research for an enterprise buyer pulls from ChatGPT's retrieval layer. If your brand is missing from the ChatGPT citation pool that compressed by twenty-one percent in six weeks, it is missing from those agents too. The pricing question is a finance problem. The citation question is a pipeline problem.
Step 2: Build internal credit-burn telemetry, even without a public rate card
Track agent runs by team, by use case, and by tool calls. You will not have a dollar number yet. You will have a structure ready the day OpenAI publishes the rate. That preparation is worth more than waiting for the rate card.
Step 3: Demand procurement-stage transparency from your AI vendors
Add a sourcing clause to every new AI agent contract: vendor must publish a unit-cost rate card before paid availability. This is not theoretical. Workspace Agents is the first widely deployed agent product to violate that norm. It will not be the last.
Step 4: Lock procurement language for the next AI agent rollout
Anthropic, Google, and Microsoft all have agent products on the roadmap. The Workspace Agents pricing pattern will recur. A standard procurement template that requires published per-unit pricing, included monthly volume, and overage rate before contract sign-off is the cheapest insurance you can put in place this quarter.
Step 5: Treat Workspace Agents as a high-priority GEO surface despite the silence
The pricing risk does not change the visibility risk. If your category prompts produce Workspace Agent runs that do not cite your brand, your buyers are reading research that excludes you. We covered the structural visibility problem in how to run an AI visibility audit. The Workspace Agents surface goes on that audit list now.
Two procurement frames, side by side
The shift in mindset is the part most teams underweight. Here is the contrast.
Traditional SaaS procurement asks:
- •What is the per-seat price?
- •What is the annual commit discount?
- •What is the renewal escalator?
- •Is there a published price page?
AI agent procurement asks:
- •What is the per-credit dollar rate?
- •What is the included monthly credit pool per seat?
- •What is the overage rate?
- •What workflows actually burn the most credits in our environment?
- •Is there a public rate card before contract sign-off?
The Workspace Agents launch is the first product where the second set of questions matters more than the first. If your procurement team is still asking the first set, the budget surprise lands in the next quarter, not this one.
Agent products price like cloud usage, not like SaaS seats. The buyer playbook has to change with them.
What to watch in the next two weeks
Three signals will tell you which way the pricing publication breaks.
A Help Center article in the next seven days. The default OpenAI pattern for delayed commercial documentation is a quiet Help Center publication. Watch help.openai.com for a new Workspace Agents Pricing or Workspace Agents Credit Math article. Probability we put at roughly twenty-five percent.
Continued opacity through May 18 to May 25. The most likely outcome based on the pattern so far. Credit usage continues to accrue against undisclosed rates. The rate card publishes only after enterprise-customer or media pressure. Probability we put at roughly sixty-five percent.
A free-period extension. Would be a credibility hit, given the May 6 cutover was publicly announced. Probability we put at roughly five percent. The remaining five percent is some hybrid: tiered seat-based pricing without full credit math, which is a way to publish something without resolving the buyer's actual question.
FAQ
What is ChatGPT Workspace Agents?
ChatGPT Workspace Agents are persistent agents inside ChatGPT Business, Enterprise, Edu, and Teachers tiers that perform multi-step research, retrieval, and workflow tasks across a workspace's connected apps. They run on credit-based pricing scaled to agent complexity, tools called, and execution time.
Is Workspace Agents free?
No. Workspace Agents was free during the rollout window, which closed on May 6, 2026. Since then it has been a paid feature with credit-based pricing. The per-credit dollar rate has not been published as of May 11, 2026.
How much do ChatGPT Workspace Agents credits cost?
OpenAI has not publicly disclosed the credit-to-dollar conversion rate, the per-action cost, the included monthly credit pool per seat, or the overage rate. Five days after paid availability went live, the only rate card is the qualitative model that credits scale with complexity, tools called, and execution time.
Should B2B SaaS teams adopt ChatGPT Workspace Agents before pricing is published?
Limited adoption is defensible. Production-scale adoption is risky without a published rate card. The safer pattern is to run controlled pilots, track usage patterns by team, and hold off on rolling agents into standardized workflows until the unit cost is known and procurement can model spend.
How should procurement handle the Workspace Agents pricing silence?
Add a contractual clause requiring a published unit-cost rate card before paid availability for every new AI agent contract. For Workspace Agents specifically, document the gap and request a written rate commitment from your OpenAI account team before scaling usage. Treat the absence of a public rate card as a procurement-policy violation and escalate accordingly.
Does this affect AI visibility and GEO strategy?
Yes. Workspace Agents are a citation surface. They pull from ChatGPT's retrieval layer when running research tasks for enterprise users. Brands missing from that retrieval layer are missing from the agent output. The pricing risk is a procurement issue. The citation risk is a pipeline issue. Both need owners.
Your buyers' agents are running research on your category today.
We benchmark your visibility across ChatGPT, Workspace Agents, Claude, Gemini, Perplexity, and Copilot, then map the gaps to the off-site sources actually feeding each surface. Most clients see measurable movement within sixty days.
Book a Discovery CallThe summary
OpenAI shipped Workspace Agents on a credit model. It started charging on May 6, 2026. It has not published the rate that connects credits to dollars. Day five of that silence is past the normal envelope by a factor of three.
For procurement, the absence of a rate card is a policy issue worth escalating now, not after the first invoice lands. For GEO and AI visibility teams, the pricing question is a distraction from the bigger one: Workspace Agents are running research on your category, and your buyers are reading that output, whether the rate card is public or not.
The right move is to run both lanes in parallel. Lock procurement language for the next agent rollout. Audit your visibility in the surface that already exists.
Continue the brief
ChatGPT Is Reading Customer Gmail. Email Is Now GEO.
OpenAI's May 5 to 6 rollout made GPT-5.5 Instant the new default and shipped Memory Sources with connected Gmail. Your email program is now a GEO surface.
Why GPT-5.5 Cites Brand Sites Less
GPT-5.5 cites brand domains in 47% of responses versus 57% under GPT-5.4. The drop comes from one query change. Here is what to do about it.
Why AI Engines Cite Same Brands but Different Sources
BrightEdge analyzed 5 AI engines across 9 verticals. Brand overlap clusters at 36-55%, source overlap spans 16-59%. The split that matters.
Framework
Learn the CITE framework behind our GEO and AEO work
See how Comprehend, Influence, Track, and Evolve turn AI visibility into an operating system.
Services
Explore our managed GEO services and AEO execution model
Audit, prompt discovery, content execution, and ongoing monitoring tied to AI search outcomes.
GEO Agency
See what a managed GEO agency should actually do
Compare real GEO operating work against generic reporting or tool-only approaches.
Audit
Start with an AI visibility audit before execution
Understand prompt coverage, recommendation gaps, source mix, and where competitors are winning.