Updated July 2, 2026 — both models live

Claude Sonnet 5 vs Fable 5

Anthropic shipped both inside 48 hours: Sonnet 5 launched June 30, 2026, and Fable 5 returned globally on July 1 after its export-control suspension. Same 1M context, a 5x price gap — here is the full side-by-side and when each one is the right call.

Quick verdict

Since July 24, 2026, Opus 5 is the default model on Claude Max. The routing ladder now has three tiers: start at Opus 5, drop to Sonnet 5 when budget or latency matters, and escalate to Fable 5 only for the top tier.

  • Opus 5 ($5 / $25 per M) — the default for most work since July 24, 2026.
  • Sonnet 5 ($2 / $10 per M, intro pricing through August 31, 2026) — for high-volume, token- or latency-sensitive work where Opus-grade quality is more than enough.
  • Fable 5 ($10 / $50 per M) — the narrow escalation for long-horizon agentic, Mythos-grade security coding, and 1M full-repo analysis.

As of August 2026: Sonnet 5's introductory pricing ends August 31, 2026. Fable 5 carries a mandatory 30-day retention and is excluded from zero-data-retention agreements.

The three-tier routing ladder

Since July 24, 2026, Opus 5 is the default on Claude Max. Route by work: Sonnet 5 for volume, Opus 5 for most requests, Fable 5 for the long-horizon extreme.

TierPrice per MBest forWhen to escalate
Claude Sonnet 5
Budget workhorse — high volume
$2 / $10 (intro, through Aug 31, 2026)High-volume coding, tool use, browser/terminal agents; latency-sensitive requestsDefault when a token or cost budget matters more than peak accuracy
Claude Opus 5
The default
$5 / $25Most routine-to-hard work — coding, agents, analysisThe default on Claude Max since July 24, 2026; escalate only when a task hits its ceiling
Claude Fable 5
The narrow escalation
$10 / $50 (30-day retention)Long-horizon agentic, Mythos-grade security coding, 1M full-repo analysisEscalate for multi-day autonomy, one-shot correctness, or whole-repo context

Prices are USD per million tokens, as of August 2026. Sonnet 5's intro price runs through August 31, 2026. Fable 5 requires a mandatory 30-day retention and is excluded from zero-data-retention agreements.

Spec-by-spec comparison

DimensionClaude Sonnet 5Claude Fable 5
TierMid-tier — the most agentic SonnetMythos-class, priced a tier above Opus
API model IDclaude-sonnet-5claude-fable-5
Intro price (in / out per M)$2 / $10 (through Aug 31, 2026)$10 / $50 (no intro discount)
Standard price$3 / $15 per M$10 / $50 per M
Context window1,000,000 tokens1,000,000 tokens
Max output128K tokens128K tokens
AvailabilityAll plans — default for Free and Pro; every major cloud and IDEPro, Max, Team, select Enterprise — included up to 50% of weekly limits through July 7, then usage credits
Best forHigh-volume coding, tool use, browser/terminal agentsHardest reasoning, multi-day autonomy, expert analysis

Both models use the tokenizer introduced with Claude Opus 4.7, so the price gap below is the real gap — there is no tokenization difference between them to adjust for. (A ~30% upward adjustment applies only when comparing against Opus 4.6 or earlier.) Sources: Anthropic announcements, June 30 – July 1, 2026 and the Claude token-counting documentation.

When to use which

Pick Claude Sonnet 5 when…

  • You run high-volume coding, refactors, or tool/browser/terminal agent loops all day
  • Cost matters and “close to Opus 4.8” quality is enough
  • You want the fast default in Claude Code, Cursor, or Copilot
  • You're pre-July-7 budgeting: Sonnet 5 stays cheap after Fable 5 moves to usage credits

Pick Claude Opus 5 when…

  • You want the default on Claude Max — most requests route here since July 24, 2026
  • You need a strong balance of quality, cost, and latency for daily coding and agents
  • You're not hitting a ceiling that only Fable 5 can clear
  • You want zero data retention out of the box

Pick Claude Fable 5 when…

  • The task is the hardest tier — huge multi-day refactors, one-shot correctness, expert judgment
  • Peak reasoning quality justifies paying 5x per token
  • You need long-horizon Mythos-class capability that Opus 5 does not cover
  • Long-horizon autonomy: Fable 5 sustains multi-day agent runs with file memory

Availability status (July 2026)

Sonnet 5 — live everywhere since June 30, 2026: Claude apps (default for Free/Pro), Claude Code, API, AWS Bedrock, Google Vertex, Microsoft Foundry, Cursor, VS Code, GitHub Copilot, OpenRouter.

Fable 5 — back worldwide since July 1, 2026 after the June 12 export-control suspension: API, Claude apps on paid plans, Claude Code, Cowork; AWS, Google Cloud, and Microsoft Foundry re-enabling. Some routine coding requests temporarily fall back to Opus 4.8.

Live Fable 5 status checker →

Sonnet 5 vs Fable 5 FAQ

Build on Claude Fable5 with your eyes open

Know the Claude Fable5 cost math, the safeguard behavior, and the right tier before you commit. Create a free account to get notified when specs, prices, or access channels change.