Anthropic shipped Claude Fable 5.1 on 1 Sep 2026 — same $10 / $50 list as Fable 5, with cache reads cut 75% to $0.25 per 1M tokens. If you already run Sonnet 5.5 for everyday work, Opus 5.5 for hard judgment, or Fable 5 for long agent jobs, the launch math is the question — not another frontier scoreboard. When does a long-context, cache-warm job make Fable cheaper in practice than your Sonnet or Opus default? When is Fable just an expensive uncached input line?
This is a use Fable when cache wins / skip Fable keep Sonnet or Opus / wait / stay Fable 5 sheet. Not a rewrite of our Claude Opus 5.5 upgrade post, and not a rematch of our Claude Sonnet 5.5 upgrade post. Full Claude vs OpenAI ladder stays on API cost for freelancers. Agent-session meters are a different layer on OpenAI Agents API budget. Here the only decision is whether claude-fable-5-1 earns this long-context job.
Affiliate disclosure: some tool or platform links may later be affiliate or referral. Prices below are from Anthropic’s introducing post, product page, and Claude Platform pricing docs as of 1 Oct 2026 (SAST) — re-check before you quote a client. Not legal advice; not a guarantee of clients, savings, or income.
What shipped 1 Sep 2026 — Fable 5.1 in plain English
On 1 Sep 2026, Anthropic published Introducing Claude Fable 5.1 and Claude Mythos 5.1. The Claude API model ID is claude-fable-5-1. Same ID shape on Bedrock (anthropic.claude-fable-5-1), Google Cloud, Microsoft Foundry, and Claude Platform on AWS — confirm each console’s exact string before you swap a production default.
Anthropic positions Fable 5.1 as the successor to Fable 5 for long-running agentic coding, multistep research, and document, spreadsheet, and slide work — same input/output prices, with cache reads at a quarter of Fable 5’s. The Fable 5.1 overview is explicit: for most workloads, start with Claude Opus 5 (family); use Fable 5.1 for demanding reasoning and long-horizon agentic work, or when your Opus evals at higher effort still fall short. Context is 1M; max output 128K; adaptive thinking always on; Claude API default effort is high; knowledge cutoff Jun 2026. Available to Pro, Max, Team, and Enterprise users, plus the Claude API and the cloud marketplaces above. Claude Mythos 5.1 is the same model under trusted-access / Project Glasswing programs only — not a freelancer self-serve flip.
Cross-link the Opus and Sonnet sheets for mid/hard everyday rungs — do not rematch those upgrade stories here.
Same $10/$50 headline, cheaper cache
Two numbers get mashed together in secondary posts. Keep them separate.
List price. On the Claude Platform pricing page and the Fable product page, Fable 5.1 is $10 input / $50 output per 1M tokens — unchanged vs Fable 5. Cache writes: $12.50 (5-minute) / $20 (1-hour). Cache reads: $0.25 per 1M — 0.025× base input. Fable 5 still reads cache at $1 (0.1×). Batch API is 50% off → $5 / $25. US-only inference is 1.1× on input/output (and related token categories) where you pin it.
“~25% typical / up to ~45% highly agentic.” Those are Anthropic’s effective-cost estimates vs Fable 5 from the cache-read cut, measured on August 2026 usage at default effort — vendor estimates, not a Mr1Tech guarantee that your next invoice drops 25–45%. Quote the USD rows, then meter one real job. Do not invent personal bill savings, ZAR conversions, or Claude Pro message quotas.
Use Fable 5.1 now — when cache economics beat Sonnet/Opus for this job
Fable earns the seat when the same large prefix keeps coming back: a long system prompt plus repo, policy pack, or research bundle reused across many turns; multi-hour Claude Code or agent loops where cache reads dominate the bill; docs→deck research packs that keep the same source bundle warm.
Absolute cache-hit dollars can look close on paper. Anthropic’s pricing footnotes put Fable 5.1 cache hits at $0.25 (0.025×), Opus 5.5 at $0.20 (0.05×), and Sonnet 5.5 at $0.20 (0.1× of a cheaper base). The catch is uncached input: Fable is $10 vs Opus $4 and Sonnet $2. Fable only wins when hit rate plus job hardness justify that premium — not because “frontier” sounds better in a Slack update.
Meter one non-secret long-context job with prompt caching on before flipping defaults. Same brief, same acceptance criteria. Log cache_read vs uncached input share, token bill, rework minutes, and wall time. If cache reads stay a tiny share of tokens, you paid Fable’s uncached line for a Sonnet-shaped job — skip Fable for that pattern. Joburg freelancers who meter Claude API in USD on a card while quoting clients in rands feel that mismatch fast; package the A/B in the quote language you already use — do not invent a rand conversion for this sheet.
Skip Fable — keep Sonnet 5.5 or Opus 5.5
Skip Fable when the job is short or cache-light: one-shot edits, small prompts, low reuse. Sonnet 5.5 at $2 / $10 remains the wallet default for well-scoped everyday coding, bug fixes, and polished docs. Open-ended architecture and judgment where the Opus sheet already said keep Opus — stay on claude-opus-5-5 at $4 / $20. Anthropic itself says start with Opus for most workloads; Fable is the long-horizon / demanding rung, not the everyday default.
Freelancer rule: if your one-job A/B does not show cache reads as a large share of tokens, do not buy Fable “because frontier.” The Opus 5.5 freelancer upgrade covers Opus judgment weeks — do not rewrite that story here. Sonnet 5.5’s mid-tier switch is live on the Claude Sonnet 5.5 freelancer upgrade sheet — cite that when you need the Sonnet decision, not this post.
Wait / stay — when not to touch the default yet
Wait when the client SOW locks claude-fable-5 or an older Opus/Sonnet ID — change order first, not a silent string swap. Wait when you have not metered cache_read vs uncached share on one job; a launch post alone is not a production flip mid-client week. Wait when the brief needs invite-only cyber or life-sciences access beyond general Fable — that is not a self-serve model-picker flip.
The Fable 5.1 overview lists real migration edges from Fable 5: forced tool use returns an error; thinking blocks bind to the model and conversation (newer accounts cannot edit earlier turns while preserving thinking the old way); earlier models cannot read Fable 5.1 thinking blocks. Using Fable means 30-day data retention by default for safety monitoring; Enterprise Frontier Safeguards (customer-held data) rolls later this fall, and eligible customers can use zero retention until then — one practical privacy check for client SOWs, not a compliance essay.
Stay on Fable 5 only if quality already ships and you refuse a one-job A/B this week. Same $10 / $50 list means “wait for a discount” is not the reason — migration caution, SOW lock, or no meter yet is.
Buy/skip matrix (sheet)
| Job | Use Fable 5.1 (cache-heavy) | Skip Fable — Sonnet 5.5 | Skip Fable — keep Opus 5.5 | Stay Fable 5 | Wait (SOW / Mythos / no meter) |
|---|---|---|---|---|---|
| Multi-hour coding agent, stable repo prefix | Yes — meter cache share | Weak if hardness needs Fable | If Opus already ships judgment | If you refuse A/B | If partner ID unconfirmed |
| Overnight research pack, same sources | Yes — warm cache | If mid-tier quality enough | If open-ended judgment dominates | If already shipping | — |
| Short bug fix / one-shot draft | No — uncached $10 hurts | Yes — wallet default | Only if judgment bottleneck | — | — |
| Architecture judgment, then Sonnet implement | Rarely for implement leg | Implement on Sonnet | Opus plans | — | If SOW locks IDs |
| Client SOW locks older model ID | Only with change order | If contracted | If contracted | Yes until SOW update | Yes — change order first |
| Biology / cyber dual-use (may fall back to Opus) | Expect fallback; not an access chase | — | Often the routed path | — | Invite programs only |
Optional printable of this free guide. Soft link until checkout. Not a “paid summary.” Not a guarantee of clients, savings, or income.
This week’s action — one non-secret long-context job
Pick one non-secret long-context job you already run on your current default. Same brief, same acceptance criteria. Run it once on that default and once on claude-fable-5-1 with prompt caching on. Compare cache_read vs uncached input share, token bill, rework minutes, and wall time. Fill one sheet row: use Fable 5.1 when cache wins / skip Fable keep Sonnet 5.5 / skip Fable keep Opus 5.5 / stay Fable 5 / wait for SOW, invite-only access, or a meter you have not run yet.
If cache share stays tiny, stay on Sonnet or Opus per the Opus and Sonnet sheets — do not panic-buy Fable. Do not change a production default mid-client week on a launch post alone.
Sources: Fable 5.1 introducing post, Fable product page, pricing, Fable 5.1 overview — re-check model IDs, cache multipliers, effort defaults, and Pro/Max/Team/Enterprise availability at paste time.