No incumbent model, no eval history, and a default to pick this week.
Route to Claude Opus 5 claude-opus-5 $5 / $25 per MTok · cache
reads $0.50
Start on Opus 5. Anthropic’s own routing guidance is to make it the default for most workloads, and at $5/$25 per MTok it is half Fable 5.1’s base rate for the same 1M context window and 128K output ceiling.
Watch for Build the eval before you build the opinion. Effort already defaults to high, so a model you never measured at a setting you never changed gives you nothing to compare an upgrade against.
The lineup as of September 2026 → Sources: Choosing a model ·Pricing
Hours-long terminal work, deep tool chains, a task the model has to hold together on its own.
Route to Claude Fable 5.1 claude-fable-5-1 $10 / $50 per MTok · cache
reads $0.25
Fable 5.1 is the documented reach for demanding reasoning and long-horizon agentic work. On Anthropic’s published numbers its margin over Opus 5 is widest exactly here — Terminal-Bench-Science 0.1 at 52.6% against 29.0%, AutomationBench at 31.4% against 26.9% — while CursorBench 3.2.0 has them within a few points.
Watch for Those are Anthropic’s scores, cited as their claims, not measurements this site reproduced. The base rate doubles to $10/$50, so the case for the switch has to come from your own runs on your own tasks.
Published benchmarks → Sources: Fable 5.1 and Mythos 5.1 ·Choosing a model
The harness works, the scores do not, and you are deciding whether to spend more per token.
Route to Claude Fable 5.1 claude-fable-5-1 $10 / $50 per MTok · cache
reads $0.25
This is the escalation Anthropic names explicitly: when evals on Opus 5 at higher effort still fall short, move to Fable 5.1.
Watch for Note the words “at higher effort.” Raise effort on Opus 5 and re-measure before you change models; effort is the cheaper knob. And check which surface you measured on — Fable 5.1 defaults to High effort in Claude Code but Medium on Claude.ai and Cowork, which is enough to make one model look like two.
Effort defaults differ by surface → Sources: Choosing a model ·Fable 5.1 and Mythos 5.1
Repo map, system prompt, tool schemas, and a growing transcript re-read on every turn.
Route to Claude Fable 5.1 claude-fable-5-1 $10 / $50 per MTok · cache
reads $0.25
Fable 5.1 bills cache reads at 2.5% of base input — $0.25 per MTok — where Opus 5 bills 10%, or $0.50. It is twice the base input price and half the cache-read price, and an agent loop is mostly cache reads.
Watch for Cache reads are one line on the bill. Fresh input and output still bill at $10/$50 against Opus 5’s $5/$25, so the crossover depends on your ratio of cached prefix to new tokens per turn. Anthropic frames the change as roughly 25% savings on typical workloads and up to about 45% on highly agentic ones; treat that as a vendor figure and run the ledger below on your own traffic.
Cache reads and the cost model → Sources: Fable 5.1 and Mythos 5.1 ·Pricing
A user is waiting, and the difference between good and best is not worth the seconds.
Route to Claude Sonnet 5 claude-sonnet-5 $2 / $10 per MTok · cache
reads $0.20
Sonnet 5 is the documented balance point between speed and intelligence, with the same 1M context and 128K output as the tiers above it at $2/$10 per MTok.
Watch for That $2/$10 was announced as introductory through 2026-08-31 and was widely expected to rise to $3/$15 on September 1. It did not; $2/$10 is now the standard rate. Any internal budget still carrying the increase is wrong in your favour — fix it anyway.
The lineup as of September 2026 → Sources: Pricing ·Choosing a model
File reads, test reruns, formatting passes — the turns that do not need planning.
Route to Claude Haiku 4.5 claude-haiku-4-5 $1 / $5 per MTok · cache
reads $0.10
Haiku 4.5 at $1/$5 is the cheap tier, and the new per-message effort beta means the mechanical turns no longer have to inherit the session’s setting even when you keep them on a larger model.
Watch for Haiku 4.5 is the odd one out in every column: a 200K context window rather than 1M, 64K output, a February 2025 knowledge cutoff, and a retirement floor of 2026-10-15 rather than deep into 2027. Keep it away from anything that depends on current library APIs, and put that date in a calendar.
The lineup as of September 2026 → Sources: Models overview ·Pricing
One model writes the plan, a cheaper one carries it out, and the handoff is code you own.
Route to Claude Fable 5.1 claude-fable-5-1 $10 / $50 per MTok · cache
reads $0.25
Fable 5.1 plans well enough to justify the split, with Sonnet 5 or Haiku 4.5 executing. The handoff is where the migration cost lives.
Watch for Earlier models cannot read Fable 5.1’s thinking blocks, so everything the planner needs to communicate has to be in ordinary output content. Forced tool use now errors as well — if you pin tool_choice to guarantee structured output, that is the most likely hard failure on migration. Editing earlier turns invalidates thinking, which breaks transcript compaction, retry-with-correction, and replay.
Breaking changes that hit harnesses → Sources: Fable 5.1 platform overview ·Claude Opus 5