TermdockCalculate Total Cost Before Assuming Savings
The CLI client, model, subscription, and API are different layers. An open-source tool does not make model requests free; plan usage, extra usage, and API bills may be separate. Establish a baseline from your own bills before changing models or task assignments. An earlier title claimed a drop from $200 to $50 per month without a verifiable measurement; that should not be treated as an expected result.
Pricing Reality: Separate Access Paths
| Cost | What to check |
|---|---|
| Fixed subscriptions | Paid plans, limits, reset periods, and whether the CLI is included |
| API and extra usage | Provider, model, input/output, caching, and actual bill |
| Other tools and environments | Additional agents, cloud compute, services, and local resources |
| Work time | Handoffs, retries, testing, and rework; record separately from the bill |
Check Anthropic's cost docs, OpenAI pricing, and GitHub Copilot plans against your own account and actual invoices. Prices and eligibility change; this article does not assume a number stays current.
A Free Client Does Not Mean a Free Model
Google's May 2026 announcement says Gemini CLI stopped serving individual free and Google AI Pro/Ultra users on June 18; Standard/Enterprise and paid API access remain supported. The Gemini CLI authentication and quota docs still list personal accounts and free quotas. The official sources conflict: verify your own account and do not assume the old 1,000 daily requests. The free-tool comparison covers other access paths.
Task Routing: Start with a Small Comparison
Choose several similar, low-risk tasks. Record the requirement, CLI/model, completion, manual correction time, and actual usage. One session can explore and another implement or review, but each additional agent may add usage. If a cheaper setting causes more rework, the per-request price alone is not enough to compare.
Hand Off and Retain Verification in Termdock
Open your installed CLI in the project workspace and give the implementer the requirement, allowed scope, and tests. When another opinion is useful, send the same requirement and diff location to a read-only reviewer. Termdock can hand tasks between terminals manually or through its CLI/API; see the agent handoff walkthrough. Record tests and usage on both sides and check each provider's bill. Termdock does not provide a unified model bill or guarantee that a handoff saves money.
A Monthly Budget Template
Record date, tool and model, authentication path, task, subscription/API cost, verification result, and reason for any retry. Add fixed subscriptions and actual extra spending over the same period. Do not add estimated session costs to an inclusive subscription a second time. Mark missing data rather than presenting an invented total.
Tracking Spending
Anthropic's Claude Code cost guide says the /usage session dollar figure is an estimate. API users should check Console usage and billing; subscribers should check plan usage rather than treating the estimate as an extra charge. Verify other providers against their own billing pages as well.
When a Higher Setting Is Worth Keeping
For security, financial, or other high-impact changes, a lower single-request cost should not displace necessary tests and human review. Compare settings against the same acceptance criteria. Lower cost may not be the better choice if errors, retries, or data risk rise.
A 30-Day Observation Plan
Record the original workflow in week one. In weeks two and three change one setting or task split, using the same acceptance criteria. In week four compare bills, quality, and human time. Consider a plan change only if the actual plan supports it and the new workflow is sufficient. Without comparable records, say "not measured" rather than claiming a fixed saving.