What changed.

Budget $20 for monthly Professional seats; the $16, $55, and $90 references require annual billing. This corrects our billing labels and does not establish a new vendor price-change date.

Small teams can now compare a two-member Team subscription. Enterprise budgets need annual seat commitments plus model usage; migrate model IDs and test workload quality before applying Opus 5.5's $4/$20 rate. Final model and rollout selection needs editorial review.

Recalculate both fixed subscriptions and included credit: Core costs less but its included credit also falls from $25 to $20. Pro's annual equivalent is $90; Agent and cloud overages remain separate. These are changes detected against our saved data, not asserted release dates.

Compare the selected Ultra access level and storage instead of a single $249.99/5 TB assumption. Consumer subscriptions do not replace Workspace team governance. U.S. reference prices require regional checkout confirmation.

Do not assume Cost mode is a no-cost baseline. Model the paid seat separately from per-user pools, routed-model prices, the third-party Token Rate, and the decision to leave on-demand use enabled. Set per-user limits before a broader agent rollout.

Use $18 only for a qualifying first-year purchase completed in the documented offer window. Do not turn the $25.20 display into a cancellation assumption: confirm the checkout and contract term before comparing it with a true month-to-month assistant or calculating savings.

Use the 350-credit path only after confirming site activation and eligibility. It can lower the cost of a tightly scoped pilot, but it is not a safe baseline for a company-wide Rovo Dev budget; otherwise model Standard at $20 per developer per month plus approved extra usage.

Use the trial to test code-review workflow, credit allocation, and billing setup, not to assume terminal or IDE fit. Budget a paid Standard validation if those surfaces are decisive, and evaluate Jira Coding Agent as a separate product path.

Recalculate new and current team budgets against the current 1,900 or 3,900-credit baseline, not the expired promotional allowance. Set organization and user spending controls before enabling usage-based overages.

Do not replace a production GPT-5.6 budget with Astra assumptions yet. Treat Astra as a separately governed high-cost pilot until access, safety controls, and workload gains are validated for the buyer's organization.

Re-run routing and unit-cost models: Luna is now one-fifth of its launch input and output price, while Sol and Terra also became cheaper for production agent workloads.

Update model allowlists and evaluation matrices to the official IDs. Use Fable 5.1 for the most demanding long-horizon work and benchmark Opus 5 as the lower-cost starting point instead of carrying forward Opus 4.8 or an invented Opus 5.1 label.

Remove the planned price step-up from forecasts. Sonnet 5 remains a materially cheaper production tier than Opus 5 and Fable 5.1 for workloads that pass quality evaluation.

Budget Work as an agentic consumption surface and validate workspace eligibility, roles, enabled apps, allowed actions, and cloud-versus-local access before rollout. Do not assume an included chat seat grants every file-creation or desktop-control workflow.

Treat the selected workspace, mount policy, and enabled connector or MCP set as the governance unit—not only the Claude seat. Pilot on approved folders with a review gate, separate usage or spend cap, and a clear escalation path for consequential actions.

Business becomes a lower fixed-seat baseline, but coding-heavy teams must separately forecast token-credit consumption. Do not use the Business seat price as the complete Codex budget.

Keep bundled core AI and autonomous execution in separate budget lines. Before scheduling work, set the spend owner, allocation, alert threshold, hard limit, and human fallback for a paused workflow.

Evaluate Grok by job surface, not as one undifferentiated chat seat. A voice-bot pilot should have an approved knowledge set, tool and transfer boundaries, recorded-call review, retention decisions, and a hard spend cap; it is not included in a Business seat.

The seat price is not the whole budget for agent-heavy teams. Start Cost mode as the baseline, then enable higher modes for a controlled group only after their output lift justifies routed-model spend.

Choose Core or Pro based on intended concurrent background work, then model Agent effort and API pass-through separately from included credits. A plan price is not a complete all-in cost ceiling.

Use the incentive to reduce onboarding cost, but do not claim annual savings from the promotional rate. Compare the steady-state Workspace price with any separate assistant or Microsoft 365 base-plan-plus-Copilot alternative.

Annual buyers should recalculate the paid baseline before comparing Synthesia with a social-video tool. The updated annual figures are still not a like-for-like comparison with a vendor that meters output differently.

A team should not forecast Zebracat spend from the headline price alone. Confirm the current checkout allowance, its unit, and whether revisions or generative work consume the same pool before approving a publishing-volume plan.

Teams should now model OpenAI as a three-tier GPT-5.6 family instead of a single premium flagship. Sol keeps GPT-5.5's flagship token price while improving agentic and computer-use performance; Terra and Luna create clearer cost-down paths for scaled assistants and subagents.

Grok is no longer only a business-packaging challenger. Engineering teams now have a materially stronger, lower-cost coding-agent pilot to compare with GPT-5.6 and Claude, though EU availability and workspace maturity still limit immediate standardization.

API-heavy agent teams have a time-limited reason to benchmark Sonnet 5 now: its launch rate undercuts GPT-5.6 Sol and Claude Opus while approaching higher-tier agentic performance. Budgets must still model the September price step-up.

Teams comparing v0 against Figma Make, Bolt, Lovable, or Replit should model v0 as a Vercel-native workflow with per-user subscriptions plus credit consumption, not as a flat app-builder subscription.

Teams comparing Claude as an agent-stack model should stop using the old 4.6 labels. Fable is now the highest-capability Claude option, Opus remains the complex coding and enterprise-work choice, Sonnet is the balance point, and Haiku is the scaled low-cost option.

Grok is still a challenger seat, but buyers should now model it around Grok 4.3 plus connector and coding-agent coverage instead of the older Grok 3 / Grok 4 Heavy framing.

App-builder buyers should model Replit around agent concurrency and Enterprise governance, not only monthly credits and collaborators.