Fable Brain

Claude 5 Model Family

5 min read

Claude 5 Model Family

The current Claude lineup as of 2026-07-07 is five pinned-snapshot models: Claude Fable 5 and Claude Mythos 5 at 50 per MTok with 1M context, Claude Opus 4.8 at 25 with 1M context, Claude Sonnet 5 at 15 with 1M context, and Claude Haiku 4.5 at 5 with 200k context.

What it is

  • The Claude 5 generation opens with Claude Fable 5 and Claude Mythos 5, announced 2026-06-09 on one shared underlying model, defining the Mythos-Class Model Tier above the Opus class.
  • Current API model IDs are claude-fable-5, claude-mythos-5, claude-opus-4-8, claude-sonnet-5, and claude-haiku-4-5-20251001 (alias claude-haiku-4-5).
  • Every Claude model ID is a pinned snapshot; dateless IDs from the 4.6 generation onward are also pinned snapshots, not evergreen pointers.
  • Four of the five are generally available; Mythos 5 is restricted to approved Project Glasswing organizations with no self-serve signup.
  • Fable Mythos 5 System Card now gives public evaluation evidence for the top shared Fable/Mythos model, especially safety, cyber, CBRN, prompt-injection, and behavioral evaluations.

The research pack does not define the formal boundary of "Claude 5 family" membership: the claude.ai capture calls Fable 5 the first Claude 5 family model, Sonnet 5 carries the 5 name, but whether Opus 4.8 and Haiku 4.5 count as family members is unstated. This note therefore covers the current lineup, not a certified family roster.

How it works

  • Price ladder per MTok input/output: Fable 5 and Mythos 5 at 50, Opus 4.8 at 25, Sonnet 5 at 15, Haiku 4.5 at 5.
  • Sonnet 5 currently runs a promotional rate; the docs note “Introductory pricing of 10 per MTok applies to Claude Sonnet 5” (https://platform.claude.com/docs/en/about-claude/models/overview).
  • Context windows: 1M tokens for Fable 5, Mythos 5, Opus 4.8, and Sonnet 5; 200k for Haiku 4.5.
  • Fable 5 and Mythos 5 add up to 128k output tokens per request, a January 2026 knowledge cutoff, and the Claude Opus 4.7 tokenizer, which produces roughly 30% more tokens than pre-4.7 models for the same text.
  • The family interlocks operationally: refused Fable 5 requests fall back to Claude Opus 4.8, and rerouted requests bill at the serving model’s prices.
  • App surfaces can also switch away from Fable 5 when safeguards fire, so logging the actual serving model matters in family-level comparisons.
  • Fable 5 and Mythos 5 are Covered Models with mandatory 30-day retention and no zero data retention; the pack documents no such designation for Opus 4.8, Sonnet 5, or Haiku 4.5.
  • The claude.ai capture of 2026-06-09 lists a different lineup: “‘claude-fable-5’, ‘claude-opus-4-8’, ‘claude-sonnet-4-6’, and ‘claude-haiku-4-5-20251001’” (System Prompt Export 2026-07, L21-23).

The corpus lineup names Sonnet 4.6 where current docs name Sonnet 5, and omits Mythos 5 entirely. Resolution: the capture is a launch-day snapshot pinned to 2026-06-09; Sonnet 5 evidently shipped after that date, and Mythos 5 is excluded from the consumer prompt because it is Glasswing-restricted. The corpus lineup is stale, not wrong for its date. CONTESTED

Knowledge cutoffs for Opus 4.8, Sonnet 5, and Haiku 4.5 are not in the research pack; only the Fable 5 and Mythos 5 cutoff (January 2026) is verified. Re-verify before citing any other cutoff.

Best practice

  • Pin exact model IDs in production configs; every current ID including the dateless ones is a snapshot, so upgrades are always explicit. EVIDENCE-BASED
  • Default agent workloads to the cheapest capable family member and escalate to Fable 5 only when the task demands it; see Model Selection for Agent Workloads. PRACTITIONER
  • Configure Fable 5 refusal fallback to Opus 4.8 so declined requests degrade inside the family instead of failing. EVIDENCE-BASED
  • Re-verify the lineup against the models overview page before citing it; the June capture went stale within a month. PRACTITIONER
  • Budget with the caching discount in mind on high-volume work; see Prompt Caching Economics before comparing raw list prices. PRACTITIONER

Pitfalls

  • Treating dateless model IDs as evergreen aliases that silently upgrade; they are pinned snapshots.
  • Copying the corpus lineup (Sonnet 4.6, no Mythos 5) into current documentation.
  • Assuming Haiku 4.5 shares the 1M window; it is the lone 200k model in the lineup.
  • Quoting Sonnet 5 at 10 as list price; that is introductory pricing over a 15 list rate.
  • Ranking the family by a single benchmark number; the system card gives public Fable/Mythos evidence, but family-wide general-purpose comparisons still require task-specific evals.

Sources

Next actions

  • Recheck the models overview page for the end of Sonnet 5 introductory pricing and any new snapshot IDs.
  • Close the family-boundary gap when Anthropic publishes a definitive Claude 5 family definition.
  • Verify knowledge cutoffs for Opus 4.8, Sonnet 5, and Haiku 4.5 before any cross-model cutoff claim.

Graph View

Backlinks