STATUS UPDATE — 8 OCTOBER 2026
Anthropic has moved past the models compared here. The current line is:
●
Claude Opus 5.5 — the frontier model
●
Claude Sonnet 5.5 — $2.00 / $10.00 per million tokens. Cache reads cut from $0.20 to $0.10 on 7 October
●
Claude Haiku 5.5 — $0.10 / $0.50 per million below 100K tokens, released 7 October
●
Claude Fable 5.1
The analysis below still describes how these model tiers differ from each other. The version numbers and prices have moved.
Current Claude pricing and benchmarks →
QUICK VERDICT — JULY 28, 2026
● Cheaper (intro pricing through Aug 31): Sonnet 5 — $2/$10/M vs Opus 5 $5/$25/M (60% cheaper input)
● Cheaper (after Aug 31): Sonnet 5 — $3/$15/M vs Opus 5 $5/$25/M (40% cheaper input)
● Better effort control: Opus 5 only — 5-level toggle
● Better for hard reasoning: Opus 5 — ARC-AGI-3 30.2% (Sonnet 5 not published)
● Better for writing and coding at lower cost: Sonnet 5 — strong output, cheaper
● Claude Max default: Opus 5 (as of July 24)
● Claude Pro: Both available — Opus 5 is now the strongest Pro model
● Intro pricing deadline: Sonnet 5 at $2/$10/M ends August 31, 2026
| Feature | Claude Opus 5 | Claude Sonnet 5 |
|---|
| Input /1M (current) | $5 | $2 (intro) |
| Input /1M (after Aug 31) | $5 | $3 |
| Effort toggle | 5-level | None |
| ARC-AGI-3 | 30.2% | Not published |
| Claude Max default | Yes (Jul 24) | No |
Claude Sonnet 5 for: Cost-sensitive high-volume workloads, writing, coding, and analysis tasks where the 5-level effort toggle is not required. Lock in usage before August 31 at $2/M if you have a high-volume use case — the intro pricing window is closing.
Claude Opus 5 for: Hard reasoning tasks that benefit from the effort toggle, long-form output (128K), Claude Max default (no settings change needed), and agentic workflows with auto-fallback replacing hard refusals. The right model once Sonnet 5 intro pricing ends and the gap narrows to 40%.
Last updated July 28, 2026. Related: Claude Opus 5 full review → · Claude Opus 5 vs Fable 5 →