Augment Code pricing: the $100 plan, usage, and real cost

The current price in one table

The Business subscription is unusual because the monthly price and included usage are both $100. For a team that consumes the entire allowance, the fixed payment behaves like a prepaid platform balance with seats, Context Engine, agents, tools, analytics, standard Cosmos compute, and security controls attached. A team that consumes little still pays the full base. A team that consumes more buys usage beyond it.

The included balance is shared by the whole workspace. A developer who runs long agent tasks and a reviewer who opens the product twice a week draw from the same pool. This is better than paying a full seat price for every occasional user, and it makes individual cost attribution less obvious. Budgeting should model the distribution of usage rather than multiply a nominal per-seat amount.

What counts as usage

The Business balance covers three broad categories: model inference, Context Engine work, and Cosmos compute. They have different cost drivers. A coding task can consume all three, so the prompt count by itself is a poor forecasting unit.

Suppose the underlying provider bills $10 for a batch of model calls. The Augment model line is $14 after the 40 percent service fee. Any Cosmos compute for that task is added separately. Context Engine consumption also draws from the shared allowance according to the product meter. The current public page does not publish a universal dollar price per retrieval query because usage can differ by product and workflow.

Model choice can change the inference portion sharply. Augment currently documents Claude, GPT, Gemini, Kimi, and Prism routing options. Its model page describes relative consumption, with lighter models below Sonnet and premium Opus or GPT options above it. The current Business pricing page translates the live bill into provider list-price dollars plus the fee, so the account dashboard should settle any conflict with older credit ratios.

Context Engine can lower model spend when it prevents broad search, repeated file reads, and repair turns. It can also add its own usage line. The net effect depends on the task. A retrieval query that saves three expensive model turns may pay for itself. A query on a small repository where grep would immediately find the file adds cost without changing the result.

Cosmos concurrency has a similar shape. The Business comparison lists standard compute and up to 50 concurrent sessions. Starting 20 workers at once increases throughput and can consume the shared balance quickly. Concurrency is a capacity limit, not a statement that those 20 sessions are included at zero marginal cost.

Why search results show other prices

Augment changed its commercial model several times as agents replaced completions and inference costs became more variable. The official blog archive preserves those announcements, which is useful history and a source of confusion.

The message model charged the same unit for a small CSS edit and a long repository task. Augment said that mismatch created unsustainable cases, including a heavy user whose underlying service cost approached $15,000 in one month on a much cheaper plan. Credits attempted to represent the different cost of models and activities. The current dollar model exposes the provider list price, a fixed 40 percent model service fee, and compute more directly.

An old subscription may have migration terms that differ from a new workspace. Existing customers should read the actual order, renewal notice, and billing dashboard before assuming the public Business offer replaced their contract immediately. New buyers should ignore fixed Indie or Max allocations unless the live checkout still presents them for that account.

Effective cost by team size

The same $100 base is expensive for a solo developer and almost negligible per seat for a full workspace. The arithmetic below divides only the fixed subscription. It excludes overage, which is likely to dominate once active agent use spreads across a team.

Even division is a reporting fiction. If five developers each run $50 of model and compute usage, a 20-person workspace consumes $250. The first $100 sits inside the plan and $150 becomes overage. The effective total is $250, or $12.50 per seat across all 20 members, while the five heavy users account for the full variable spend.

A sound forecast uses user cohorts. Separate daily agent users, occasional agent users, reviewers, managers, and Slack-only users. Run a two-week pilot, capture spend by product and model, normalize for working days, add expected concurrency, then include a margin for larger tasks and new adoption. Recalculate after the first month because successful tools create more usage than pilots.

The 50-seat inclusion changes procurement behavior. A platform team can invite reviewers and occasional users without immediately adding a per-seat bill. Access still has administrative and security cost. Remove departed users, define repository scopes, and review permissions even when the financial meter does not punish an extra seat.

Augment versus Cursor and Claude Code pricing

For one person, Cursor Pro or Claude Pro has the lower committed cost. For five people, Augment's fixed base equals $20 each before overage, which matches common individual entry plans. At 20 or 50 people, Augment's seat cost is far lower and the question becomes variable usage. Cursor Teams at 50 seats is $2,000 per month before any applicable usage beyond plan terms. Claude Team Standard is $1,000 annually billed per month at the $20 rate, again subject to its usage windows.

Feature equivalence is limited. Cursor includes an AI-first editor and predictive Tab. Claude Pro includes the broader Claude product and a terminal agent using Anthropic models. Augment includes a context service, multi-model agent, Cosmos, review, and shared team usage. Buying the cheapest nominal seat can be costly if it misses the capability the team actually needs.

Continuum is a free application around provider agents and does not sell retrieval or model usage as part of the base workbench. It uses Claude, ChatGPT, Cursor, and other subscriptions or keys the user already owns, and optional hosted inference is a separate choice. That makes its pricing additive to provider plans rather than a direct replacement for Augment's included usage.

Budget scenarios

The included $100 makes a tiny proof of concept possible under the Business plan, while a realistic team pilot may need top-ups. Set a dollar ceiling, restrict the first cohort, and choose tasks with known human baselines. Record both Augment spend and any provider subscription that remains in parallel.

A successful retrieval layer can reduce model tokens and engineering time, yet savings appear in different ledgers. The Augment dashboard shows service consumption. Developer search time and review wait live elsewhere. Decide before the pilot which operational metric has to move and how much one hour of that delay costs.

Example month for a 20-person team

Assume four daily agent users each generate $90 of underlying model inference, six occasional users generate $20 each, and the rest use review or Slack lightly. Underlying model spend is $480. Applying the 40 percent service fee produces $672 before Context Engine and compute. If those two services add $128, total measured usage is $800. The first $100 is included in the Business payment and $700 is overage, so the cash total remains $800 rather than $900. This example uses simplified list-price dollars and no taxes or contract discounts.

Now compare the result with the workflow it replaced. If Augment removes a $400 review tool and saves ten engineering hours worth $100 each, an $800 service month has a strong economic case. If the same team keeps every old subscription and saves no measurable time, the low $5 fixed seat arithmetic is irrelevant. Variable usage makes outcome accounting more important than seat accounting.

Run a downside case too. Double the heavy-user activity, use a premium model for half the tasks, and add several long Cosmos workers. Confirm that budget alerts arrive before the shared balance is exhausted and that administrators can stop overage. A forecast that includes only average days will fail during migrations, incidents, release weeks, and initial adoption spikes.

Keep a separate line for duplicated spend during rollout. Teams commonly retain Cursor, Claude, code review, and CI automation while Augment is being evaluated. That overlap is sensible during a reversible pilot and expensive when it becomes the permanent default. Give every retained tool a date, owner, and exit condition so the new platform has to earn consolidation rather than merely join the stack.

How to avoid a surprise bill

1. Use the billing dashboard as source of record. Plan examples and model ratios are estimates; account consumption controls the invoice.
2. Set workspace budgets before broad rollout. Confirm whether overage pauses, warns, tops up, or bills in arrears under the actual order.
3. Separate model and compute. A cheap model can still sit inside a long-running compute workflow.
4. Prefer the least expensive model that passes the task. Use premium models for work whose failure cost justifies the multiplier.
5. Measure Context Engine net cost. Compare retrieval usage with model turns, tokens, and repair work saved.
6. Shut down idle concurrency. A session limit does not make active compute free.
7. Export usage by team and workflow. Shared pools hide the heavy-user tail in a simple per-seat average.
8. Review historical plan language at renewal. Migration, rollover, refund, and top-up terms may differ from the public page.

Augment's general credit policy says unused credits expire at the end of the order term unless the order says otherwise, while the current pricing FAQ says standalone top-ups remain valid for 12 months from purchase. Those statements cover different buckets. Ask sales to identify which balance expires when, then retain the answer with the order form.

Questions people ask

Keep reading