Choosing among the 12 Claude model ids

First question: pool lane or dedicated lane

This decision splits the twelve before any of the others, because it moves price by 4× on the ids where you have a choice. Five Claude ids are pool-lane at a 0.1× multiplier: claude-opus-5, claude-opus-4-8, claude-opus-4-6, claude-sonnet-5, claude-sonnet-4-6. Seven are stable-* at 0.4×.

A stable-* id is a dedicated lane served on our own key rather than the shared account pool, and it does not participate in pool failover. Same weights either way — the multiplier pays for the path, not for a different model.

Four models exist on both lanes, so the choice is real for them: Opus 5, Opus 4.8, Sonnet 5, Sonnet 4.6. For the rest it is made for you. claude-opus-4-6 exists only on the pool lane, and stable-claude-haiku-4-5, stable-claude-fable-5 and stable-claude-fable-5-1 exist only as dedicated lanes.

Second question: does 200K fit

Nine of the twelve carry a 1,000,000-token context and a 128,000-token output ceiling: both Fable ids, all three Opus ids and their stable twins, claude-sonnet-5 and stable-claude-sonnet-5. Three are served at 200,000 with a 64,000-token output ceiling: claude-sonnet-4-6, stable-claude-sonnet-4-6 and stable-claude-haiku-4-5.

This is a fit question, not a quality question. Look at the input tokens your requests actually carry and the length of the responses you actually need. If neither approaches those ceilings, the larger window buys you nothing and can be dropped from the criteria entirely.

Note that context does not track price. claude-sonnet-5 gives the full million at the same $3/$15 official rate as the 200,000-token claude-sonnet-4-6, and both bill at $0.30 and $1.50 per million.

Third question: what each id costs you

Billed rates, official rate multiplied by the id’s own multiplier, per million input and output tokens. claude-sonnet-5 and claude-sonnet-4-6: $0.30 / $1.50. stable-claude-haiku-4-5: $0.40 / $2.00. claude-opus-5, claude-opus-4-8 and claude-opus-4-6: $0.50 / $2.50. stable-claude-sonnet-5: $0.80 / $4.00. stable-claude-sonnet-4-6: $1.20 / $6.00. stable-claude-opus-5 and stable-claude-opus-4-8: $2.00 / $10.00. stable-claude-fable-5 and stable-claude-fable-5-1: $4.00 / $20.00.

Two things in that list surprise people. The pool-lane Sonnets are the low number, below the small model — stable-claude-haiku-4-5 is 1.33× them, because a 0.4× multiplier on a $1/$5 official rate lands above a 0.1× multiplier on $3/$15. And stable-claude-sonnet-5 is the one stable lane priced off a lower official rate than its pool twin, $2/$10 against $3/$15.

Output is 5× input on every Claude id in the catalog, without exception. So the ratio of tokens you read to tokens you generate decides how much these differences matter — a summarisation workload and a code-generation workload rank the same list differently.

curl -sS https://token-share.app/v1/messages \
  -H "x-api-key: $TOKEN_SHARE_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Review this diff for correctness bugs."}]
  }' | jq '.usage'

# claude-sonnet-5 bills $0.30 per 1M input and $1.50 per 1M output.
# cost = input_tokens / 1e6 * 0.30 + output_tokens / 1e6 * 1.50

Fourth question: thinking, and then measure

Eleven of the twelve list text, tools and thinking. stable-claude-haiku-4-5 lists text and tools only, so it is out if a thinking budget is part of your design and in contention otherwise. Every Claude id here answers on /v1/messages, so tool use and the request shape are common ground.

Those four questions are lookups, and they usually leave two or three ids standing. What separates the survivors is quality on your tasks, and no table can rank that — the catalog records capacity and price, not behaviour.

So run them. Same endpoint, same body, one field changed: put your evaluation set through each finalist and compare the failures rather than the wins. Where the price is identical — claude-opus-5 against claude-opus-4-8, or claude-sonnet-5 against claude-sonnet-4-6 — the comparison costs you nothing but a duplicate run, and it answers a question this page deliberately does not.

Common questions