Tools/Model Selector

Model Selector

(Pricing data as of December 2025)

What Model Should I Use?

A five-step workflow to choose the right model tier for your use case.

Step 1: What are you building?

Choose the workflow that best matches your product.

Use Case

This sets workflow complexity, reasoning depth, and typical model usage.

Description: Tools that summarize, classify, and respond to email or messaging threads.

Examples: Superhuman, Gmail Help Me Write, Outlook Copilot

Primary content type: Short text & messages

Typical token size per request: Input tokens: 2,000 · Output tokens: 1,000

Step 2: How often will this be used?

Describe how frequently a typical user relies on this workflow.

Usage Level

Uses Per Day (Per User)

Think about how many emails or messages a typical user would want help with each day (for example, 5, 20, or 50).

Steps per request (internal model calls): 2

Model calls per user per day: 40

Step 3: What kind of model do you want?

Control the quality vs. cost tradeoff.

Model Family

Choose the level of intelligence and reliability.

Performance Level

Value balances cost, High maximizes capability.

Mid-tier Value — Balanced capability for most production workflows.

Fit for this workflow

Good fit

Good fit — Good fit — better quality if tone and nuance really matter to you.

Capability & Cost Profile

Reasoning

3.0 / 5

Relative cost

3.0×

Hallucination risk

3.5 / 5

Long-context strength

3.0 / 5

Step 4: What will this cost?

Estimate per-user and total costs based on your usage.

Daily active users (DAU)

How many users do you expect will run this workflow each day?

Cost summary

Costs are based on your selected model tier: Mid-tier Value

Cost per user / month

$1.08

Total monthly cost (all users)

$540

TOKEN PRICING USED

Input price per 1M tokens: $0.20

Output price per 1M tokens: $0.50

PER-TASK COST

Cost per request: $0.00

Cost per 1,000 requests: $0.90

Step 5: Our recommendation for this workflow

See our suggested configuration for this workflow.

Summary

Good fit

Email Assistant using Mid-tier Value at Typical usage.

This is a good fit — not always the absolute cheapest, but a solid choice when quality and nuance matter.

Good fit — better quality if tone and nuance really matter to you.

Why this tier works:

Quality vs cost: uses Mid-tier · Value with the fit score 4/5 for this workflow.
Costs are reasonable for this workflow and scale well with higher usage.
Hallucination risk and long-context strength are already factored into this recommendation via the fit matrix.

Example models for this tier:

Llama 3.1 8B Instruct Turbo
Llama 4 Scout
Mistral Small 3

Example workflows this workflow is used for:

Onboarding and follow-up email sequences
Reminder and notification campaigns
Internal summary or announcement emails

Our default starting tier for this workflow is Open-sourceValue. Use it if you want a simple, safe baseline without tuning.

Compare nearby tiers

See how adjacent model tiers compare to your current choice.

Open-sourceHighGood fit

Good fit

Cheaper than your current tier.

Good fit — cheap and capable if you want a bit more headroom than the baseline.

Mid-tierValueGood fit

Good fit (your current tier)

Same cost band as your current tier.

Good fit — better quality if tone and nuance really matter to you.

Mid-tierHighOverkill / tradeoff

Works with tradeoffs

More expensive than your current tier.

Overkill — premium quality that most task flows don't actually need.

Advanced (for the curious)

Uses per user / day20

Internal calls per workflow2

Calls user / day (C)40

Input tokens / request2,000

Output tokens / request1,000

Model pricing reference (actual token costs)

Input and output prices per 1M tokens used in these calculations, using median prices by tier from the Token Pricing dataset.

Model family	Performance	Input $ / 1M tokens	Output $ / 1M tokens
Frontier	High	$2.50	$6.20
Frontier	Value	$0.65	$1.68
Mid-tier	High	$0.60	$1.50
Mid-tier	Value	$0.20	$0.50
Open-source	High	$0.20	$0.80
Open-source	Value	$0.06	$0.10

1) Cost estimation methodology: Estimated costs are calculated based on the number of workflow uses, the number of steps per workflow, average input and output token usage per step, and published input and output token prices. A "step" represents a single model invocation.
(Uses × Steps × Input tokens × Input token price) + (Uses × Steps × Output tokens × Output token price)

2) Scope and assumptions: Estimates reflect inference-related token costs only and exclude infrastructure overhead, retries, tooling, and non-model execution costs. Figures are intended for comparative decision-making rather than precise billing forecasts.