Snapshots

Stage by stage costs are known; the monthly total is not

Being built
  • Plays cost 18 cents each on Pro and 83 cents on Opus, median on Arata.
  • Every stage has a cap, from a 45 dollar build to a 30 dollar learnings upload.
  • A first build on Pro is proposed at about 18 dollars; on Opus about 80.
  • No brand has a measured monthly total, and jam calls have no dollar meter.
  • Drafts are free to the team; Studio makes and finals show a price first.
6.40Cozy, 65 Plays on Flash
Your call

Measure one real month for Arata, Nutrova and Cozy by adding AI dollars to the spend file?

AI and models · 6.7

Cost per brand

DraftLast checked Mon Oct 26Owner Param

Answer

What one brand costs Brew in models and outside data, stage by stage, from the read to Plays, drafts, Studio makes and finals, with every cap that holds it and every figure that has been measured; the monthly total is not measured yet, so it is a proposal.

Today

Prices the meter uses

Model $ per million tokens in, out Source
claude-opus-5-5 4, 20 brew/ai.js:8
claude-sonnet-5-5 2, 10 brew/ai.js:8
gemini-3.5-flash 0.5, 3 brew/gemini-client.js:16
gemini-3.1-pro-preview 2, 12 brew/gemini-client.js:16
Web search $0.01 on Claude, $0.014 a grounded query on Gemini brew/ai.js:13 brew/gemini-client.js:17

These are list prices before Google credits brew/gemini-client.js:15. Today Vertex is paid by credits: $20,763 of $25,000 left on 16 Oct memory brew-gemini-provider.

Stage by stage

Stage Measured Cap Source
Read, the first look at a store Not measured as one figure treg day cap $40 across all brands; Apify $10 a brand a month brew/tools/prod-env.sh:9 brew/shelf.js:69
Plays, about 60 to 65 a brand, all kept Pro median $0.18 a Play on Arata; Opus median $0.83; Flash, Cozy, 65 Plays for $6.40 BUILD_CAP_USD $45 a build; $150 for a brand marked Pro; DAILY_BUILD_CAP_USD $250 across brands memory keep-all-moves memory brew-model-compare-2026-10-20 brew/rewrite-pot.js:19 brew/build.js:113-115 brew/tools/prod-env.sh:9
Play overhead A run costs about 1.5 times the Play's own call, with search, check and fix brew/rewrite-pot.js:24
Before the lean pipeline $1.51 a Play, most of it about 42,000 thinking tokens memory brew-lean-pipeline
Keeping Plays fresh Estimate $0.10 a Play on Flash, $0.40 on Pro, $0.60 on Claude Weekly pot a brand: $10 Flash, $40 Pro, $50 Claude; BRAND_WEEK_CAP_USD overrides brew/rewrite-pot.js:19-21 brew/rewrite-pot.js:71
First connected account $40 a run, CONNECT_CAP_USD brew/build.js:1070 memory brew-connect-refresh
A learnings upload $30, LEARN_CAP_USD brew/build.js:1075
Connected re-read rewrites $30 a brand a run, weekly, LIVE_CAP_USD brew/build.js:1076 memory brew-move-freshness
Site check refresh $60 by hand, $10 automatic brew/build.js:1069
Watching AI answers check about $0.50 a weekly run; viral finder $0.015 to $0.04 treg a weekly run; Brand Book enrichment about $0.01 Listening $0.60 a run; pulse $3 a run; enrichment $0.50 brew/live-watch.js:23 memory brew-viral-remakes memory brew-brand-book brew/listen-watch.js:684 brew/market-pulse.js:318
Outside data in a month Estimate about $10 of treg a brand Apify $10 a brand a month memory treg-tool-catalog brew/shelf.js:69
Drafts, automatic Picture $0.04, about $0.039 on Flash; Veo $0.15 a second; voice $0.002 Daily units across brands: text 1,500, image 1,500, video 60; a brand's share is a third, video 20 a brand; 40 pictures a day kept for the team brew/data/engines.json brew/deploy.sh:21 brew/budget.js:12-15 brew/tools/prod-env.sh:9 brew/server.js:564-565
Studio makes, on a yes About $0.28 a static ad; on Arata, 23 Oct: product shot $0.10, preset $0.10, Kling 5 s $0.40, Seedance UGC 10 s $3.02 Every make shows its price and needs confirmUsd memory brew-jam-v2
Finals, on a yes Image 2.01 credits; reel 45.071 credits; Arata viral reel 80.8 credits Credits a day: 300 across brands, 100 a brand; Sloosh's own daily limit, raised to 200 on 21 Oct memory brew-sloosh-finals memory brew-viral-remakes brew/budget.js:13-15
Jam and chat No dollar meter Daily text units only brew/budget.js:12-15
One-time Style example pictures, 43 for $1.64, shared by all brands memory brew-setup-v2

Where spend is recorded

  • AI calls: meter.byTag per server, and the build's day total in _brew/spend-day.json brew/ai.js:51 brew/build.js:120-124.
  • treg and Apify: per brand and month in brands/<slug>/spend/<YYYY-MM>.json brew/treg.js:121 memory brew-apify-watch.
  • Daily units: _brew/usage-day.json, shared across instances brew/budget.js:1-5.

Tomorrow

Item Version Tag
Keep all Plays; cut the cost of each one, never the count v1 Stays memory keep-all-moves
One cost line per brand per month, all sources, in dollars, on the ops page v1 New
Jam calls metered in dollars v1 New
A monthly ceiling per brand that pauses background work and says so v2 New
Cost per brand set per plan once pricing is decided: free read, 14-day trial, plans with credits v3 New, parked until after the design partnership memory brew-pricing-parked
Sloosh credits shown as dollars to the team v1 Changes

How we do it

  1. Read a brand's outside-data spend for a month:
    gsutil cat gs://brew-prototype/brands/<slug>/spend/2026-10.json | jq '.usd, (.rows | group_by(.tag) | map({tag: .[0].tag, usd: (map(.usd) | add)}))'
    On Cloud SQL, the same rows are objects under the same path memory brew-storage.
  2. Before a model switch, raise the caps that follow it: BUILD_CAP_USD, BRAND_WEEK_CAP_USD memory brew-pro-switch-and-flush.
  3. Never run paid generations while building; ask before a paid probe memory no-spend-while-building.
  4. New paid work gets a cap in env, a share per brand, and a tag that books it to the brand.

Numbers

Metric Definition Value Status
Cost per Play Median dollars of a Play's own call $0.18 Pro, $0.83 Opus, about $0.10 Flash Measured memory brew-model-compare-2026-10-20 brew/rewrite-pot.js:19
First build Every Play written once Proposed: about $18 on Pro, from 65 Plays at $0.18 and 1.5 overhead; under the $45 cap Proposed
First build on Opus same Proposed: about $80 at $0.83 and 1.5 overhead, over the $45 cap, so the cap must rise first Proposed
Monthly running cost a brand Plays fresh plus watching plus outside data plus drafts Proposed: about $180 at most before drafts, from 4 weeks of the Pro pot at $40, about $10 of treg, up to $10 of Apify and about $2 of AI answers checks; drafts add up to $20 a day at the brand's picture share; not measured Proposed
Monthly running cost, measured Sum of a brand's spend files and its AI tags not measured Proposed
Cost per Studio make Dollars per take $0.10 to $3.02 on Arata Measured memory brew-jam-v2

Open

  1. Measure one real month. No brand has a measured monthly total. Proposal: add AI dollars per brand to the spend file, then report October and November for Arata, Nutrova and Cozy.
  2. What is a Sloosh credit in dollars? Proposal: write the Pro plan price into Finance and convert finals on the ops page.
  3. Read cost. The first read has no single figure. Proposal: tag every read call reader-<slug> and sum it on the next new brand.
  4. Cap for Plays on Opus. Proposal: if Plays go back to Opus, set BUILD_CAP_USD to $90 for that deploy.
Sources
  • brew origin/main b91e364: build.js:113-115 and 1060-1080, rewrite-pot.js:14-24 and 71, budget.js:12-15, deploy.sh:21 and 41, tools/prod-env.sh:9, treg.js:45 and 82 and 121, apify.js:40, shelf.js:69, listen-watch.js:684, market-pulse.js:318, live-watch.js:23, server.js:564-565, data/engines.json, data/higgsfield-routes.json, ai.js:8, gemini-client.js:16-17
  • memory brew-model-compare-2026-10-20, brew-lean-pipeline, brew-pro-switch-and-flush, brew-move-freshness, brew-connect-refresh, brew-jam-v2, brew-sloosh-finals, brew-viral-remakes, brew-apify-watch, treg-tool-catalog, brew-brand-book, brew-setup-v2, keep-all-moves, brew-gemini-provider, brew-pricing-parked, brew-storage, no-spend-while-building

Source file content/ai-and-models/cost-per-brand.md · For agents: Markdown · JSON