BMStart lesson 1

7 models · official prices checked 12 Oct 2026

Which AI model should I use?

Short answer: a clear brief matters more than the model. But if you need to choose, this page compares every current Claude and GPT model using official data.

Answer 3 questions Start with AIM

Compare

Compare at a glance

Pick a company, then switch between cost and what each model is best for.

Input: what you sendOutput: what it writesUS$ per million tokens

Each step on the scale is 10× the price. Official list prices, checked 12 October 2026.

The lineup

Every model, one card

Tier, best use and price, from each company’s official documentation.

Anthropic4

Most capable

Claude Fable 5.1

The hardest reasoning and long-horizon agent work

in
$10
out
$50
Released 1 Sept 20267 min
Recommended start

Claude Opus 5.5

Long-running agentic work and complex knowledge work

in
$4
out
$20
Released 22 Sept 202610 min
Balanced

Claude Sonnet 5.5

The best mix of speed and intelligence for everyday work

in
$2
out
$10
Released 28 Sept 20268 min
Fastest

Claude Haiku 5.5

High-volume classification, extraction, routing and sub-agents

in
$0.10
out
$0.50
Released 7 Oct 20267 min

OpenAI3

Flagship

GPT-6 Astra

The hardest reasoning and long agentic tasks

in
$10
out
$50
Released 3 Sept 202611 min
Balanced

GPT-6.1 Sol

Near-flagship coding, computer use and professional work at lower cost

in
$2
out
$10
Released 29 Sept 202611 min
Fastest

GPT-6 Luna

Focused, high-volume tasks at the lowest cost

in
$0.10
out
$0.50
Released 22 Sept 2026No guide yet

Not sure?

Answer three plain questions and get one suggestion per company.

Go to the picker

US$ per million tokens

Picker

Which one should I use?

Three questions, no jargon. Suggestions follow how each company describes its own models.

Answer three questions. The suggestion updates as you tap.

1How hard is the task?
2How often will you run it?
3Which tools do you use?

Suggested

AnthropicClaude Sonnet 5.5Balanced$2 / $10Read the guide →
OpenAIGPT-6.1 SolBalanced$2 / $10Read the guide →

For everyday drafting and analysis, the balanced tier gives the best mix of speed, quality and cost.

US$ per million tokens, input / output. A rule of thumb based on each company’s own description of its models. A clear brief matters more than the model.

Shared habits

What both vendors agree on

Anthropic’s and OpenAI’s official prompting advice has converged. These six habits work with every model above.

  1. 1. Clear and direct

    Write as if to a brilliant colleague who has none of your context.

  2. 2. Goals over recipes

    Give the goal and constraints; current reasoning models do better without over-detailed step-by-step instructions.

  3. 3. Define "done"

    State the deliverable, the format and how to verify it before the model starts.

  4. 4. Structure with tags

    Separate instructions, context and material with headings or XML-style tags.

  5. 5. Show examples

    A few varied examples often teach more than a paragraph of description.

  6. 6. Ease off hard rules

    Current models follow instructions very closely; rules in capital letters and rigid recipes now over-constrain them.

All model data checked against official documentation on 12 October 2026. Prices and lineups change often; confirm before making a purchasing decision.