distribute.you
The bet
Every AI vendor now sells the same model family in tiers. Flash, the cheapest. Pro, the middle. Frontier, the most expensive. Pro costs about 5x Flash per token, and everyone has a reflex about it: the pricier model must convert better, so the question is whether it converts enough better to pay for itself.
The test
We wrote 100,000 cold emails with both tiers, on real campaigns for real clients, and counted two things per dollar spent: how many people clicked the link, and how many wrote back with interest.
The twist
Both tiers won. Just not the same match.
Flash buys the click 6x cheaper. Pro buys the reply 8x cheaper. The premium does not pay for itself. It pays for a different outcome.
The rule
Why
A click is cheap to earn: a decent email with a link, and some people click. A reply is expensive to earn: someone has to read, believe and type. Across all 100,000 emails a positive reply was 22x rarer than a click. The stronger model earns its keep exactly where the bar is higher.
We have not tested the Frontier tier yet. At 5 to 10x the Pro price, the reply would have to get a lot cheaper still. We will publish that when we have it.
What it means if we run your campaigns
You never pick a model. Each campaign on distribute.you runs on the workflow with the best measured cost for what that campaign is buying, across every client, and the pick moves when the numbers move. You pay the budget the campaign spent, nothing else, and you watch the cost of each positive reply land on your dashboard.
Start with your website URL. The first 30 USD are on us.
Method
- Scope: sales cold email outreach only. Journalist outreach, PR pitches and every other channel we run are excluded.
- Window: April 14 to September 8, 2026, every client account. Production data from the distribute.you platform, read on September 8, 2026.
- Volume: 95,661 emails to 33,243 people, $6,716 of platform spend, 698 website visits, 31 positive replies. 20 workflows (a workflow is one prompt and one model); workflows configured but never run are excluded.
- Tiers: Flash, the cheapest model of a family (Gemini Flash, Claude Haiku, DeepSeek V4 Flash, GLM Flash, Kimi K2.6). Pro, the middle (Gemini Pro, Claude Opus, DeepSeek V4 Pro, GLM 5.3, Kimi K3). Frontier, the most expensive (Claude Fable, Astra), not yet run in production.
- What ran: Flash tier, 4 workflows on Gemini 3.5 Flash-Lite and Gemini 3.5 to 3.8 Flash, 34,506 emails, $1,369. Pro tier, 12 workflows on Gemini 3.1 Pro, 57,650 emails, $4,869. DeepSeek V4 Pro and GLM 5.3 (Pro tier, 3,505 emails, $478) are too early to rank and are not in the charts.
- Spend: everything the workflow cost before per-account discounts: lead sourcing, enrichment, sending and tokens. Not tokens alone.
- Website visit: a recipient clicked the tracked link. Counted once per person.
- Positive reply: a recipient wrote back with interest (a question, a request for details, a meeting). Classified automatically, reviewed by a person on the account. Negative replies, out-of-office and unsubscribes do not count.
- Template check: the tiers did not run the same prompts evenly. On the prompt family that asks for a click only, Flash still bought the visit for less ($3 vs $8) and Pro still bought the reply for less ($256 vs $1,369).
- Limits: not a randomised experiment. Our optimiser assigns workflows by measured cost, so Pro workflows ran more on reply-seeking campaigns. 31 positive replies is a small count. We report counts and no confidence intervals.
Per-workflow data and list prices
Spend in USD, rounded to the dollar. CPWV is cost per website visit, CPPR is cost per positive reply, left blank under two replies.
| Workflow | Tier | Model | Emails | Spend | Visits | Replies | CPWV | CPPR |
|---|---|---|---|---|---|---|---|---|
| Lithium | Pro | Gemini 3.1 Pro | 25,378 | 2,231 | 78 | 19 | 29 | 117 |
| Rampart | Flash | Gemini Flash-Lite | 19,396 | 618 | 222 | 0 | 3 | |
| Pelican | Flash | Gemini Flash | 9,458 | 408 | 100 | 0 | 4 | |
| Permafrost | Pro | Gemini 3.1 Pro | 9,008 | 597 | 87 | 2 | 7 | 299 |
| Legato | Pro | Gemini 3.1 Pro | 8,231 | 390 | 90 | 0 | 4 | |
| Azalea | Pro | Gemini 3.1 Pro | 5,493 | 593 | 4 | 4 | 148 | 148 |
| Osprey | Flash | Gemini Flash-Lite | 4,302 | 231 | 86 | 1 | 3 | |
| Ballad | Pro | Gemini 3.1 Pro | 3,163 | 395 | 0 | 4 | 99 | |
| Cerulean | Pro | DeepSeek V4 Pro | 1,868 | 336 | 0 | 1 | ||
| Trailblazer | Pro | Gemini 3.1 Pro | 1,357 | 84 | 0 | 0 | ||
| Dawn | Flash | Gemini Flash-Lite | 1,350 | 113 | 15 | 0 | 8 | |
| Vector | Pro | Gemini 3.1 Pro | 1,241 | 156 | 14 | 0 | 11 | |
| Tango | Pro | Gemini 3.1 Pro | 881 | 85 | 0 | 0 | ||
| Moraine | Pro | Gemini 3.1 Pro | 875 | 84 | 0 | 0 | ||
| Colchis | Pro | Gemini 3.1 Pro | 831 | 85 | 0 | 0 | ||
| Helium | Pro | Gemini 3.1 Pro | 697 | 85 | 0 | 0 | ||
| Rudder | Pro | DeepSeek V4 Pro | 579 | 47 | 0 | 0 | ||
| Allegro | Pro | DeepSeek V4 Pro | 531 | 47 | 0 | 0 | ||
| Nobelium | Pro | GLM-5.2 / 5.3 | 527 | 47 | 2 | 0 | 24 | |
| Wisteria | Pro | Gemini 3.1 Pro | 495 | 84 | 0 | 0 |
List prices per million tokens, USD, standard context, paid tier, read from each vendor's own page on the date shown.
| Model | Tier | Input | Output | Source, date read |
|---|---|---|---|---|
| Gemini 3.5 Flash-Lite | Flash | 0.30 | 2.50 | ai.google.dev, 2026-09-08 |
| Gemini 3.8 Flash | Flash | 0.75 (1.50 from 2027-01-01) | 3.75 (7.50 from 2027-01-01) | ai.google.dev, 2026-09-08 |
| Gemini 3.1 Pro Preview | Pro | 2.00 | 12.00 | ai.google.dev, 2026-09-08 |
| DeepSeek V4 Flash | Flash | 0.22 off-peak, 0.44 peak | 0.66 off-peak, 1.32 peak | api-docs.deepseek.com, 2026-08-15 |
| DeepSeek V4 Pro | Pro | 0.66 off-peak, 1.32 peak | 1.98 off-peak, 3.96 peak | api-docs.deepseek.com, 2026-08-15 |
| GLM-5.3-Flash | Flash | 0.15 (0.075 promotional) | 0.50 (0.25 promotional) | docs.z.ai, 2026-09-08 |
| GLM-5.3 | Pro | 1.40 | 4.40 | docs.z.ai, 2026-09-08 |
| Kimi K2.6 | Flash | 0.95 | 4.00 | platform.kimi.ai, 2026-08-16 |
| Kimi K3 | Pro | 3.00 | 15.00 | platform.kimi.ai, 2026-08-16 |