Why most agents default to the wrong Claude tier

Anthropic’s current stack is four tiers, and the choice is an economics problem, not a benchmark one. Most teams default to Opus for everything because it tops the benchmarks, or stay on Sonnet out of habit - and pay the wrong rate for most steps of an agent workflow either way. Fable 5 (claude-fable-5, $10 / $50 per million input/output tokens) is the frontier escalation tier for the hardest reasoning and longest-horizon work. Opus 5 (claude-opus-5, $5 / $25) is the orchestrator - near Fable intelligence at half Fable’s token rate, same price as Opus 4.8. Sonnet 5 (claude-sonnet-5, $2 / $10 through 2026-08-31, then $3 / $15) is the default daily driver. Haiku 4.5 (claude-haiku-4-5, $1 / $5) is the worker you fan out across parallel sub-agents. Match the tier to the job and you stop paying Fable or Opus rates for work Haiku finishes faster.

Definitions

Fable 5 - Anthropic’s frontier tier (claude-fable-5), highest widely available capability, built for the hardest reasoning and longest-horizon agentic work. Ships with safety classifiers that can decline a request as an HTTP 200 with stop_reason: "refusal" - every integration needs refusal handling, not just Opus-style error handling. See Claude Fable 5: refusals are an API contract, not an error.

Opus 5 - Anthropic’s orchestrator-tier model (claude-opus-5): near-Fable agentic quality at Opus price, replacing Opus 4.8 for long-horizon work. 1M-token context, up to 128K output tokens. Adaptive thinking; effort defaults to high on the Claude API and Claude Code. See Claude Opus 5 closes on Fable at Opus price.

Sonnet 5 - the default worker tier (claude-sonnet-5), replacing Sonnet 4.6. Same 1M-token context, adaptive thinking, and on Anthropic’s launch benchmarks it strictly improves on Sonnet 4.6 - at Low-to-Medium effort it matches Opus quality at a lower per-token price. See Claude Sonnet 5 closes the gap to Opus without the Opus bill.

Haiku 4.5 - the fast, cheap tier for high-volume and latency-sensitive work. 200K-token context, 64K max output. Built to run as a sub-agent under a more capable orchestrator.

The tiers at a glance

Prices, context windows, and output caps are from Anthropic’s model documentation, current as of 2026-07-27. The effort parameter is Opus- and Sonnet-tier only - it returns an error on Haiku 4.5, so Haiku is configured through the older thinking controls, not effort.

Pick by job, not by benchmark

Benchmark deltas between tiers are real but they are not the decision. The decision is which role the model plays in your workflow and what that role costs per token. Four roles:

The expensive mistake is running every step at Fable or Opus rates. The other expensive mistake is forcing a deep, long-horizon migration through a tier that loses the thread halfway. Tier selection is the lever; the sections below are when to pull it.

Encoded as a config your scripts and sub-agent launchers can read:

FRONTIER=claude-fable-5        # rare, hardest-step escalation only
ORCHESTRATOR=claude-opus-5     # long-horizon, high autonomy (was claude-opus-4-8)
DEFAULT=claude-sonnet-5        # daily driver (was claude-sonnet-4-6)
WORKER=claude-haiku-4-5        # parallel sub-agents

When to reach for Fable 5

Fable 5 is still the top widely available rung, but Opus 5 narrowed the gap - Anthropic positions Opus 5 as coming close to Fable intelligence at half the price. Reach for Fable only for the single hardest step in a run: the one piece of reasoning or the one long-horizon call where Opus 5 isn’t enough. Every call needs to branch on stop_reason: "refusal", since Fable’s safety classifiers can decline a request as a successful HTTP 200 rather than an error. Wire a fallback to Opus 5 (or enable Anthropic’s beta automatic fallbacks) rather than stopping at Opus 4.8. At $10/$50 - double Opus’s rate - and with that added integration surface, Fable 5 is an escalation tier, not a default. Full detail: Claude Fable 5: refusals are an API contract, not an error.

When to reach for Opus 5

Reach for Opus 5 when the task is long-horizon, hard to fully specify up front, and expensive to get wrong: a multi-day refactor, a codebase-scale migration, or an orchestrator that delegates to a fleet of workers and has to keep the plan straight across hundreds of tool calls. Swap claude-opus-4-8 for claude-opus-5 at the same $5/$25 - same bill, better agentic runs.

The capability case is concrete. On Anthropic’s launch numbers, Opus 5 more than doubles Opus 4.8 on Frontier-Bench v0.1 at a lower cost per task, lands within 0.5% of Fable 5’s peak on CursorBench 3.2 at max effort and half the cost per task, and on OSWorld 2.0 beats Fable 5’s best result at just over a third of the cost. That is self-verifying, multi-step work - a cheaper tier that compounds small errors into a wrong result loses here.

Two operational notes:

• Give it the whole spec up front and run at high effort. Opus 5’s long-horizon strength comes from reasoning more at each step. A complete first turn plus effort: "high" or "xhigh" produces more efficient and more accurate runs than drip-feeding context.
• Fast mode is an Opus-tier lever. When latency matters, Opus 5 runs the same model at roughly 2.5× output speed at twice the base price ($10/$50) - a knob Sonnet and Haiku don’t have.

Opus 5 also ships lighter safety classifiers than Fable - cybersecurity-focused, with ~85% fewer interventions. Flagged requests can fall back to Opus 4.8 by default; beta automatic fallbacks on the API cover Opus 5 and Fable 5. Full detail: Claude Opus 5 closes on Fable at Opus price.

When Sonnet 5 is the right default

Sonnet 5 is the model you should reach for first and escalate from only with a reason. It replaced Sonnet 4.6 as the default worker tier on 2026-06-30, carries the same 1M-token context as Opus, and on Anthropic’s launch benchmarks it strictly improves on Sonnet 4.6 at every effort level on BrowseComp (agentic search) and OSWorld-Verified (computer use) - a real upgrade, not a rename.

Pricing is intro-then-standard: $2 / $10 per million input/output tokens through 2026-08-31, then $3 / $15. Anthropic set the intro price so the switch from Sonnet 4.6 is roughly cost-neutral despite an updated tokenizer that produces up to ~30% more tokens for the same text. At Low and Medium effort, Sonnet 5 delivers near-Opus quality at a lower per-token cost. At High and XHigh effort, that cost-per-quality advantage narrows or disappears - escalate to Opus 5 rather than pushing Sonnet 5 to its limits. Full detail: Claude Sonnet 5 closes the gap to Opus without the Opus bill.

Stay on Sonnet 5 for interactive coding, document Q&A, and design-heavy frontend work. Escalate to Opus 5 only when the task is long-horizon or autonomy-critical - and remember that a bigger window is not what makes the harder tier worth it. The discipline that actually keeps agents reliable is context engineering, and it applies at every tier.

When Haiku 4.5 wins

Haiku 4.5 is the sub-agent economy. At $1 / $5 it costs about half of Sonnet 5’s intro rate (a third once standard pricing resumes on 2026-09-01), and on Anthropic’s numbers it lands 73.3% on SWE-bench Verified while matching Sonnet 4-class coding and computer-use quality at up to 4–5× the speed of Sonnet 4.5. That combination - cheap, fast, capable enough - is exactly what a parallel worker needs.

Use it for real-time chat, high-volume classification and extraction, and any architecture that fans many scoped tasks out at once. The one constraint to design around: Haiku’s context window is 200K, not 1M, so keep each worker’s job small and self-contained rather than handing it a sprawling history.

The orchestrator–worker pattern

The tiers compose. The pattern that gets the most out of the stack is an Opus 5 orchestrator that delegates scoped sub-tasks to Haiku 4.5 workers, with Sonnet 5 as the solo default when no orchestration is needed, and Fable 5 called directly for the rare step that needs it. That is worker-tier routing.

Symptom - a single big model reads a 30K-token API reference or a pile of source files, and that clutter stays in context for the rest of the run, dragging quality down. Cause - heavy reads done in the main loop persist on every subsequent turn. Solution - push the read into a sub-agent that runs in its own window and returns only the conclusion. Spawn those workers on Haiku and you get the isolation and the cheapest possible token rate for the grunt work. See subagent context isolation.

This is why the per-token spread matters more than the benchmark spread: in a fan-out architecture, the worker tier runs far more tokens than the orchestrator, so a cheaper worker is a cheaper bill on the dominant cost - and that spread now runs as high as 10:1 (Fable 5 to Haiku 4.5) across the full stack, not just within the three workhorse tiers.

Bottom line

There is no single best Claude model - there is a best model per role. Default to Sonnet 5, escalate to Opus 5 for long-horizon and high-autonomy work, reserve Fable 5 for the single hardest step in a run, and delegate the parallel grunt work to Haiku 4.5. Pick by the job and the token bill, not the leaderboard, and the stack pays for itself. Tiering is one lever on that bill - prompt caching and batch processing are the two with a fixed, guaranteed discount. When the choice is openness and hosting rather than Claude tier, use Open-weights ≠ open-source for agent stacks.