neomanex
Claude Fable 5 logo

Claude Fable 5

Anthropic · Released Jun 2026

View pricingCompare models
Conditional

Anthropic's Mythos-class frontier model — the strongest publicly-available Claude yet, leading every major benchmark at launch. The verdict stays CONDITIONAL on two live constraints: credit-only pricing at $10/$50 per 1M tokens with no subscription path, and a safety classifier that over-triggers on routine coding tasks, silently routing them to an Opus 4.8 fallback. Reach for Fable 5 when the capability delta justifies the cost; for routine work, Sonnet 5 is the more predictable alternative.

Is it right for you?

Good for

  • Long-horizon autonomous coding — compresses months of engineering into days (Stripe: 50M-line migration in 1 day vs 2-month team effort)
  • Complex analytical reasoning — Hebbia Finance Benchmark #1, senior-research-scientist-grade hypothesis generation
  • Vision-intensive tasks — state-of-the-art chart/table interpretation, rebuilds web apps from screenshots, beats Pokémon FireRed vision-only
  • Agentic workflows with persistent memory — autonomously executes multi-turn tasks with file-based memory, self-corrects from notes

Not good for

  • Cybersecurity and biology queries — safety classifiers fall back to Opus 4.8; legitimate security/life-sciences work hits the safeguard wall
  • Cost-sensitive high-volume production — at $50/1M output, batch processing costs compound aggressively vs GPT-5.5 at $30 or Opus 4.8 at $25
  • Regulated environments with data-residency constraints — 30-day mandatory data retention for Mythos-class models is a compliance burden

How it performs by task

Code generation

Excellent

Class-leading: 80.3% SWE-bench Pro, #1 FrontierCode, Stripe-validated at production scale

Long-context reasoning

Excellent

Stays focused across millions of tokens; persistent memory improves game-playing 3x over Opus 4.8

Knowledge work

Excellent

Top finance/analytics benchmark scores; strong on legal redlining, trading analysis, root-cause reasoning

Vision

Excellent

New state-of-the-art — extracts precise numbers from scientific figures, vision-only game completion

Scientific research

Excellent

Produces novel molecular biology hypotheses (80% scientist-preferred over Opus-class); 10x drug-design acceleration

Cybersecurity (offensive)

Fair

Mythos 5 scores 78% ExploitBench, but Fable 5 blocked by classifiers — falls back to Opus 4.8

Pricing

Input

$10 / 1M tokens

Output

$50 / 1M tokens

Context

1M ctx, 128K out; Max plans at 50% limits

View full pricing

Benchmarks

BenchmarkScoreSource
SWE-bench Verified95.0% Source
SWE-bench Pro80.3% Source
FrontierCode Diamond29.3% Source
Terminal-Bench 2.188.0% Source
Humanity's Last Exam (tools)64.5% Source
GDPval-AA1932 Elo Source
LiveCodeBench89.78% Source

Verdict history

Jul 15, 2026
rewritten for brevity (content-OS density pilot) — Copy edit to new 2-4 sentence / 55-90 word verdict_summary standard. All detail preserved in existing structured children (benchmarks[], strengths[], verdict_audiences[], task_strengths[], pricing_context).

Sources

Verification log

  • Profile— Updated

    Automated agent

    Re-verification of the access finding first logged 2026-09-07, confirmed independently against the live page today. Anthropic's plan table lists Claude Fable as: Free not available, Pro usage credits, Max 5x and Max 20x 50% of weekly limits, Team not available, Enterprise (self-serve and sales-assisted) available. A subscription path therefore EXISTS, and the stored verdict_summary claim 'credit-only pricing at $10/$50 per 1M tokens with no subscription path' is false. API rates re-verified UNCHANGED on the platform rate card: $10 in / $50 out per 1M, cache reads $1 per 1M; Fable 5 carries no limited-availability tag, unlike Mythos 5 and Mythos 5.1 which link to anthropic.com/glasswing. ONE field changed this run: pricing_url moved from https://www.anthropic.com/pricing (not fetched, unverified) to the platform rate card actually read. The structured children were already corrected on 2026-09-07. THE SUMMARY WAS NOT, AND THIS IS THE FINDING: the 2026-09-07 pass staged the rewrite on a draft named 'claude-fable-5-max-plan-access-correction'; that draft EXISTS (post_create returned 409 duplicate slug this run) but was never published, so the verdict never moved and the entity has continued to state the false half in the field readers quote while its own pricing_context and audience rows state the true half. No duplicate draft created. The entity_verdict payload for the EXISTING draft is staged in this run's return. Rating held at conditional on the surviving safety-classifier condition. NOT ESTABLISHED, deliberately left open: the plan table says 'Fable' with no version number while the API card lists Fable 5 and Fable 5.1 separately, so which version a Max subscriber reaches is unresolved and was not guessed.

  • Profile— Updated

    Automated agent

    Resolves the carried access trigger. API rates re-verified UNCHANGED at $10 in / $50 out per 1M with cache reads at $1 per 1M (platform.claude.com pricing table). ACCESS CHANGED against stored: the verdict summary asserts 'credit-only pricing ... with no subscription path', but Anthropic's live plan table lists Claude Fable as Free: not available, Pro: usage credits, Max 5x and Max 20x: 50% of weekly limits. A Max subscription path exists; the credit-only description holds for Pro only. This confirms our own W29 published record (Fable 5 added to Max/Team from Jul 20) against a live primary, closing a stale-self-authored-prose drift open since 2026-07-15. Added a Max-subscriber for_whom audience and a Free-tier not_for_whom, refreshed pricing_context to note Max at 50% limits. Fable 5 carries NO 'limited availability' tag on the pricing table (unlike Mythos 5 and Mythos 5.1). Rating held at conditional on the surviving safety-classifier condition; verdict summary rewrite staged for the announcing draft 'claude-fable-5-max-plan-access-correction'.

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

  • Status— Updated

    Automated agent

    Summary re-dated to Jul 7 2026: added Jul 1 return, safety-router over-blocking (silent Opus 4.8 fallback), and the Jul 7 credit cliff (credit-only $10/$50 from Jul 8, no auto fallback, no return timeline). Verdict stays conditional; condition restated as credit-only access + over-triggering safety router.

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

    Pricing unchanged: $10/$50 per 1M tokens. Anthropic pricing page STILL shows 'Claude Fable 5 is unavailable.' CryptoBriefing reports ban lift but independent tracker isfable5back.com confirms API still suspended. Ban lift NOT reflected on primary source (Anthropic pricing page).

  • Pricing— No changes

    Imported at launch

  • Profile— No changes

    Imported at launch

How we evaluate