Source: ai-research/anthropic-claude-fable-mythos-5-1-announcement-2026-09-01.md — Anthropic’s launch page (anthropic.com/claude-fable-and-mythos-5-1; the page shows “September 2026”, and the system card is dated 2026-09-01), fetched 2026-09-29. Claude Code facts from v2.1.257; migration facts from the claude-api skill update (anthropics/skills #1704). The October 2026 update cites its podcast sources inline; the Ben AI section (2026-10-28) cites its three videos. Anthropic’s own launch and model-selection videos (added 2026-10-30) are cited in their section.
Claude Fable 5.1 (claude-fable-5-1) and Claude Mythos 5.1 (claude-mythos-5-1) are the same model with different safeguards: Fable 5.1 is generally available, while Mythos 5.1 is available only through trusted-access programs for cybersecurity and life-sciences work. Anthropic frames the release around the three customer complaints that capped Fable 5’s adoption — price, data retention, and safeguards — and addresses each directly. Three weeks later Opus 5.5 claimed Fable-5.1-level work on most tasks at 40% of the per-token price, so Fable 5.1’s case now rests on the hardest long-horizon work.
Key Takeaways
- Same list price, much cheaper cache reads. 50 output per MTok (unchanged from Fable 5); cache reads $0.25 per MTok, 75% less. Anthropic’s estimate from four weeks of real August usage: ~25% cheaper for typical workloads, up to ~45% for highly agentic, context-heavy work where cache reads dominate the bill.
- Zero data retention is back for eligible customers, with a longer-term replacement. Enterprise Frontier Safeguards (EFS) stores data “in cloud infrastructure controlled entirely by the customer,” with any human review done by the customer by default. It rolls out in phases “beginning later this fall” across Claude Code, Claude Enterprise, the Claude Platform, Bedrock, Google’s Agent Platform, and Microsoft Foundry. Until then, eligible customers can use Fable 5.1 (and Fable 5) with zero data retention.
- Cyber safeguards fire far less. 60% fewer cyber false positives, and Fable 5.1 may now be used to discover software vulnerabilities (not to develop exploits). Claude Code users should see “around 60% fewer interventions per session.” Penetration testing, exploit generation, and binary-based vulnerability scanning still redirect to Opus models. Biology safeguards for Fable 5 and 5.1 now fire 85% less often on benign elementary biology and medical questions.
- Default effort differs by surface: High in Claude Code, Medium in Claude Cowork and Claude.ai. Anthropic says Fable 5.1 at Low or Medium matches or beats Fable 5 “at a much lower cost.” Its model-selection video puts a number on it: on CursorBench, Fable 5.1 at medium “produced a similar result to 5 at max effort for about a fifth of the cost.”
- Anti-distillation: new API accounts can no longer edit history and keep the thinking. For accounts created from launch day onward, manually editing Claude’s prior context in a multi-turn conversation while preserving its prior thinking is blocked. Existing accounts were not affected at launch, “though it will apply to all users with future model releases.” Opus 5.5 extended the same mechanism, preserved thinking, to accounts created on or after 2026-08-31.
- Watermarked output (EU AI Act). Models released after 2026-08-02 carry an invisible watermark under the EU Code of Practice on Transparency of AI-Generated Content, which Anthropic signed in July 2026. A detection API is in private preview for regulators, media, researchers, and obligated enterprises. See Claude Output Watermarking.
- Mythos 5.1 is identical but gated. Available through the Cyber Verification Program (Mythos-class access “in the near future”) and the Life Sciences Verification Program, built with the US government. At launch it was limited to a set of US organizations. Claude Security (codebase vulnerability scanning) now runs on Mythos 5.1 — see Claude Security.
Benchmarks (vendor-reported)
| Benchmark | Fable 5.1 | Fable 5 | Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | 29.0% | 22.4% |
| Terminal-Bench 4.0 | 55.8% (Mythos 5.1: 60.9%) | 42.0% | 52.3% | 37.3% |
| GDPval-AA v2 (Elo) | 1853 | 1723 | 1824 | 1711 |
| OSWorld 2.0 partial / strict | 77.9% / 41.7% | 72.9% / 36.1% | 75.4% / 39.6% | — |
| Humanity’s Last Exam no tools / tools | 60.9% / 65.0% | 57.8% / 63.8% | 56.6% / 63.6% | — |
| AutomationBench | 31.4% | 17.1% | 26.9% | 19.6% |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% | 67.2% |
- The Fable/Mythos gap on Terminal-Bench 4.0 (55.8 vs 60.9) is the safeguard cost. Anthropic: the two are the same model, and “the gap between them reflects the tasks on which our earlier, less precise cyber safeguards intervened.” It expects the gap to shrink with the new safeguards.
- Evaluated with production safeguards on: where they intervened, OSWorld and AutomationBench scored those tasks as zero, and other flagged tasks were completed by Opus 4.8 (cyber) or Opus 5 (biology).
- These numbers use different benchmark versions from the Opus 5.5 table (GDPval-AA v2 here vs v2.1 there; CursorBench 3.2.0 vs 4.0), so the two tables are not directly comparable.
What testers reported (vendor-selected)
- Root-cause debugging. Millennium: Fable 5.1 found the cause of a rare crash (about one in a million runs) that its engineers had not explained in four to five years. It disassembled a vendor library, matched it against the core dump, and traced the crash to a bug in that library.
- Long unattended runs. One tester: a 38-hour unattended ML run that diagnosed a prior result as a label artifact, corrected it, and ran six parallel experiments overnight. Another: “it keeps its own records, reprioritizes as things change, and picks up where it left off.”
- Economics changed routing. Devin’s team: “We’re moving our Opus 5 traffic in Devin to Claude Fable 5.1 on launch day… with the new cache read pricing a Fable-class model is finally economical for the workloads we’d kept on Opus, starting with code review.” Another tester: “Fable-level intelligence, Opus-level price, Sonnet-speed… about twice as fast as Opus 5 and used half as many tokens.”
- Documents and decks. One tester reported the best PowerPoint decks of any model it had tested. A contract-redlining benchmark rose from 47.9 to 57.0, and FrontierFinance from 49.2% to 55.9%, credited to going “straight to the call transcript” instead of secondary coverage.
Science results (Mythos 5.1)
- Protein binders: with open-source design and folding tools, Mythos 5.1’s designs had binding affinities 10× higher than the best Adaptyv Bio competition entries on three targets, and a hit rate of nearly 50% across 12 targets (10–15% is typical), confirmed by two external labs.
- Venus elevation map: Fable 5.1 trained a network on 30-year-old Magellan radar data to map a third of Venus at 2–3 km detail (was 10–20 km), released under Creative Commons.
- GPU kernels: custom kernels and caching sped up seven open-source protein and genomics models by up to 2.5× with identical outputs, cutting estimated GPU costs for genome-wide analyses by 30–60%.
Field reports, first month (creators and press, not controlled tests)
- Subagents inherit Fable by default, and that is where the bill goes. Nathaniel Whittemore (The AI Daily Brief): Fable 5.1’s default is to use “the most advanced model as the coordinator, but also … for all the sub agents,” so a routine research task that spins up five or ten subagents can use “a major percentage” of a usage cap (
raw/7_Ways_How_We_Use_AI_Is_Changing.md). Leo Laporte’s fix: tell it “if you create a sub agent, don’t use Fable 5.1. Use Opus or Sonnet,” since subagents are usually “collecting stuff or using tools” (Intelligent Machines,raw/Infinite_Slop_-_The_End_of_Apps_Meet_the_Agentic_OS_Built_for_Your_Phone.md). See Subagents for pinning a subagent’smodel. - The 25% saving did not show up on one independent harness. Matt Wolfe, reading Artificial Analysis’s cost-per-task chart in launch week: Fable 5.1 was the most expensive model per task at 3.14 for Fable 5 (
raw/AI_News_-_The_Most_Insane_Week_So_Far_This_Year.md). Anthropic’s estimate assumes cache-read-heavy workloads; a single-shot benchmark with little cache reuse would not see it. - Low effort for drafts, extra for the finish. Nate B Jones ran the same valuation task at low and extra effort. Low produced a finished workbook and deck but “no sources sheet… no check sheet”; extra added linked sources and questions “that could change the investment decision.” His suggested pipeline: build the first version on Fable low, have GPT-5.6 Sol check the structure, then finish on Fable extra (
raw/Everyone_s_Testing_Claude_Fable_5.1_On_Code._It_Made_Me_A_37-Second_Film..md). He also found “Claude’s limits still feel much tighter than OpenAI’s,” though 5.1 “went way farther than 5 inside those limits.” - Computer use is the gap testers route around. In a head-to-head with GPT-6 Astra, Jones uses Fable to flesh out work but goes to Astra in Codex for computer use: “the speed at which that model can use the computer, it’s night and day” (
raw/The_Race_to_Done_-_Fable_5.1_vs_GPT-6_Astra._Who_Wins.md).
Ben AI field reports (added 2026-10-28)
Three videos from Ben AI, who uses Claude for business work rather than code. Single runs; costs and counts are his own readings.
Where he uses Fable 5.1 (raw/5_Actual_Use_Cases_for_Claude_Fable_5.1_For_Professionals.md, https://www.youtube.com/watch?v=uzDH5ZpzyRg, 2026-09-04): only for complex end-to-end tasks worth the spend, after 10 to 15 minutes writing the prompt.
- Research. From Claude’s library of customer case studies, it spun up four subagents, analysed 40 stories, pulled out 260 tactics and returned a video outline with links per tip in about 10 minutes. He calls it “80% ready” to record.
- Design. It one-shotted a landing-page redesign that followed a reference site, including its hover animations. His advice: give it references, because iterating is where the tokens go.
- Data analysis. Two months of support data (346 calls, 15 Q&As, 305 tickets) became a dashboard of the 16 most common issues, a prioritised product list and a customer profile.
- Plain explanations. With an “ELI5” skill he credits to Anthropic (not verified here), Fable 5.1’s diagram of his company’s second brain was simpler and more readable than Opus 5’s from the same prompt.
- Reading a second brain. Given an open prompt about his business over the past year, Fable 5.1 read 225 files; Opus 5 read three.
The subagent instruction, priced (raw/Paste_This_Into_Claude_Never_Hit_a_Token_Limit_Again-Jr-jyTL2MYI.md, https://www.youtube.com/watch?v=Jr-jyTL2MYI, 2026-09-14). This puts numbers on the subagent bullet in the field reports above:
- A customer-intelligence report used 67 subagents, all on Fable 5.1. He estimates 80% of the tokens went to subagents and the report cost close to $350.
- The same prompt plus “When using sub-agents, always run them on Sonnet 5” produced an “almost identical” report for $44, about 87% less. ^[inferred: percentage computed from his two figures; he says “80 to 85%”]
- His safer version also forces the delegation: “When you do research or read the transcripts, always use sub agents that run on Sonnet 5.”
- Correction: Ben says “Anthropic’s own article said that Fable is best as an orchestration model and to use Sonnet or Opus 5” for the reading work. The saved launch page has a tester saying they would “absolutely use Claude Fable 5.1 as the orchestrator,” but no Anthropic advice to put subagents on Sonnet or Opus. The Opus 5 prompting guide advises capping delegation, not choosing a cheaper subagent model. The advice is his, and Leo Laporte’s above.
Against Opus 5.5 (raw/I_Tested_Opus_5.5_vs_Fable_5.1_on_7_Real_Use_Cases_Not_Even_Close.md, 2026-09-24): Opus 5.5 beat Fable 5.1 on five of seven tasks, including the design task above, at about half the cost. Fable 5.1 kept the two context-heavy jobs: it read 555 support calls to Opus’s 248 and about 200 second-brain documents to Opus’s 50, and both reports were deeper. Table in Which Claude Model for Which Job.
October 2026 update
Added 2026-10-27 from The AI Daily Brief episodes fetched 2026-10-21.
Cyber Verification Program now has three tiers (raw/The_Best_New_AI_Products_and_the_Key_Trends_They_Represent.md). At launch the program promised Mythos-class access “in the near future.” The October version opens Mythos to security professionals, together with “versions of opus and sonnet with reduced safety guardrails”:
| Tier | What it allows | Who can get it | Data retention |
|---|---|---|---|
| Defense access | Defensive work: incident response, malware reverse engineering, vulnerability analysis | Security professionals who apply | Zero data retention available |
| Red team access | Limited offensive work such as penetration testing, still with real-time blocks on actions “that could cause physical harm or mass disruption” | Enterprise customers only, after enhanced vetting | Zero data retention available |
| Specialized access | Fewest guardrails, for testing high-risk systems such as flight control, power grids and “financial plumbing” | Requires government approval | Mandatory data retention, the only tier with it |
- Per the episode, specialized access “will replace and expand Project Glasswing”, the program that gave Mythos to a few dozen companies after its April launch.
- How to use it: a security team that needs Mythos or lower-guardrail Opus or Sonnet should apply for Defense access. Red team access needs an enterprise contract.
Fable 5.5 sightings (rumour, unconfirmed). The AI Daily Brief relays X posts (Chubby and others) saying some users’ queries are “being routed to Fable 5.5”, a model with “an updated knowledge cutoff and some slick new design taste” (raw/What_the_Best_Business_AI_Users_Are_Doing_Different.md). A later episode says Fable prompts and some Opus 5.5 code-animation prompts were “routed to a new model” (raw/How_Muse_Wonder_Shifts_the_AI_Landscape.md). Andrew Curran reads Anthropic’s raised revenue outlook as confidence in “the Wizard Fable 5.5” (raw/How_the_Latest_AI_Model_Jump_is_Disrupting_Business_Models.md). Anthropic has not announced it in any source here. Brock Mesarich (“Claude Fable 5.5 is Coming + More AI News”, https://www.youtube.com/watch?v=8htNqGGxlDo, 2026-10-16, raw/Claude_Fable_5.5_is_Coming_+_More_AI_News.md) adds a sighting of his own: Fable 5.5 appeared on 2026-10-13 in Microsoft Azure’s model list, marked generally available. He notes it is the only 5.5 model not yet out (Haiku 5.5 shipped 2026-10-07) and that Anthropic has given no date, price or benchmarks. The Azure listing was not checked for this article. His usage note: on the $200 Max plan he uses Opus 5.5 “constantly” and “basically never” passes half his weekly limit, so his open question is whether Fable 5.5 will be worth Fable’s 2.5x price and faster limit burn.
Anthropic’s launch videos and routing advice (added 2026-10-30)
Source: Anthropic’s official YouTube channel (@claude). Transcripts in raw/ under each title; upload dates and descriptions in ai-research/anthropic-youtube-model-launch-video-descriptions-2026-10-30.md. Five of the six demo shorts have no narration, so what they show comes from Anthropic’s video descriptions.
Fable 5.1 demos (uploaded 2026-09-01, launch day). Each shows long work across several sources, with a person approving the result:
- Debugging across the whole stack (https://www.youtube.com/watch?v=jwztQLH76is, 1:30). Anthropic: Fable 5.1 “is particularly strong on long-running engineering work” and fixes “the root cause of an issue rather than the symptoms.” It works through months of vehicle data, support tickets and several codebases at once to find a bug the team could not reproduce internally. It proposes a cause; the developer confirms it, decides what is safe to ship and approves the fix.
- The ops review in Slack (https://www.youtube.com/watch?v=G3vwVsh9RtU, 1:26). A team builds a leadership ops review in Slack with Claude Tag. Fable 5.1 builds the deck from the team’s inputs, flags a number that does not reconcile, and reworks the calculation once someone weighs in. The team reviews the deck before it goes to leadership.
- The forecast, overnight (https://www.youtube.com/watch?v=S9IJ1GgAAxE, 1:14). One person at a B2B company maintains a revenue forecast with contract and consumption revenue. Claude updates it overnight, “running unattended on our API.” In the morning the analyst asks how accurate past forecasts were, and Claude backtests itself. The analyst approves before anything goes out.
All three descriptions end with a human sign-off, and the Slack and forecast descriptions both say Fable 5.1 checks the numbers “as it goes.” Anthropic is selling it for unattended, multi-source work that a person reviews at the end.
Three “Claude codes” demos from the same day. Their titles and descriptions do not name a model. They are listed here because they were uploaded on 2026-09-01, Fable 5.1’s launch day, with the three Fable shorts.
- Flight data into “look behind you” (https://www.youtube.com/watch?v=NkJPtSLLDLs, 0:58). A Mac app: you type your address and sketch the trees and rooftops around your yard; it watches live air traffic, works out the physics (including how long the sound takes to reach you) and counts you down aloud. A plane hidden behind your house gets “listen”, not “look.” The only narrated short; its voice says, for example, “It’s a Delta Airbus A330-300, descending through 2,600 ft at 200 knots… from Rome to New York, landing in about 5 minutes.” Data: crowd-sourced ADS-B from adsb.lol and adsb.fi, Open-Meteo elevation, and Apple’s on-device frameworks for maps and voice.
- Turn-by-turn directions for the Moon (https://www.youtube.com/watch?v=KcRcbHxeoKk, 0:37). A navigation app built from scratch that plans drives over NASA terrain data (LOLA), steering around craters too steep to enter and across shallow ones, and explains every turn. It runs on a hand-built 3D globe in the browser “with no libraries.”
- A watercolor engine (https://www.youtube.com/watch?v=3TJ9vWl0Ewo, 0:49). A simulator built from scratch that computes how water moves, how pigment settles and how light mixes colour, live in the browser with no libraries.
Anthropic’s routing advice (“Which Claude model should you use?”, raw/Which_Claude_model_should_you_use.md, https://www.youtube.com/watch?v=71-8fJIGi34, 2026-09-18, 3:45):
- Judge cost per task, not price per token. Model, effort and cache decide the cost. “A more intelligent model often finishes a task with less work.” The video repeats the 75% cache-read cut and the ~25% (typical) to ~45% (long agentic) savings from the launch page.
- Open-ended work goes to Fable 5.1, routine work stays on Opus or Sonnet. Open-ended means deep research, complex analysis and long-running agent work; routine means summarising a document, drafting an email or pulling data from a report. On Fable, “start with medium and raise it as needed.” Fable 5.1 defaults to High in Claude Code, so following that advice there means lowering it first.
- Admins get three controls: model entitlements, effort caps and default model and effort, set for everyone or per role. Full notes in Which Claude Model for Which Job.
AI Search hands-on tests and complaints (added 2026-10-30)
Source: raw/Claude_Fable_5.1_is_savage.md (AI Search, https://www.youtube.com/watch?v=RmlP_nqOhsk, uploaded 2026-09-03, about 35 min; sponsored by Higgsfield). Single runs in Claude Code at “ultra code” effort. The presenter is openly hostile to Anthropic’s marketing in this video; his opinions are labelled as such.
What worked
- Floor plan to 3D interiors: three textured three.js designs that matched the uploaded floor plan’s layout, in two prompts. “This has very good spatial understanding.”
- From-scratch ray tracer (no libraries): a sphere, cube and pyramid on an infinite ocean with working sliders for materials, waves, foam, sun and clouds, in one prompt and 30–40 minutes.
- Earnings video: it chose HyperFrames on its own, recovered from Gemini TTS errors, and produced a narrated comparison of five big-tech quarters; he rates its design above GLM 5.3’s and GPT’s.
- Matcha ad via the Higgsfield CLI: it learned the CLI unaided and made six clips plus music, but the audio overlapped because it did not notice the clips already had sound.
What did not
- Blender X-wing via MCP: “quite basic”; he preferred Opus 5’s. “In terms of creating 3D assets or designing video games, front end, etc., it feels like Opus 5 is actually better than Fable 5.1.”
- EDM track in his DAW: competent structure and a mastering chain, but “a lot of fuzz.” Hidden-frog test failed; 0 of 6 brain-tumour scans.
- Medical research prompts fell back to Opus 5. Two neutral prompts (chronic myeloid leukaemia drivers; amyloid beta and tau in Alzheimer’s) showed “switched to Opus 5,” and retrying with Fable 5.1 reverted again. The tumour-scan prompt did not switch. This matches the fallback design described in the benchmark notes above.
- Usage limits: one ultra-code prompt hit the 5-hour limit; the video “took roughly 15 hours” of testing, mostly waiting for resets.
His claims about cost and marketing (relayed, partly checked)
- Artificial Analysis cost per task 3.14 for Fable 5, and $8,500 to run its full evaluation; he calls Anthropic’s “25% cheaper” claim “blatantly lying.” The same AA figure is already recorded above from Matt Wolfe; Anthropic’s estimate assumes cache-heavy workloads.
- Highest AA Intelligence Index (66 vs GPT-5.6 Sol 61) but no confidence intervals; slowest time to first token (about three times GPT-5.6 Sol, ten times GLM 5.3); more hallucination than Fable 5, Kimi K3 and GLM 5.3 on AA’s Omniscience; #1 on Arena WebDev and LiveBench.
- Watermark: “all the responses generated by Claude models have a hidden text watermark.” The wiki’s record is narrower: models released after 2026-08-02 carry an EU AI Act watermark (see Claude Output Watermarking).
- Plans: he says Fable 5.1 is not available on Pro, only Max ($100+/month) or the API (not verified here), and points to the Max-plan usage lawsuit covered in Kahn v. Anthropic and a community note calling the September 25% weekly-limit raise a net cut against an expiring promotion.
- His conclusion: use Fable 5.1 “as a last resort” for hard non-cyber, non-bio coding or reasoning; he personally would not subscribe.
Superseded by Fable 5.2 (added 2026-11-11)
Sources: Anthropic’s Fable 5.2 launch page (ai-research/anthropic-claude-fable-mythos-5-2-announcement-2026-11-11.md), the Help Center’s “Claude Fable models on your plan” (ai-research/anthropic-support-claude-fable-models-on-your-plan-2026-11-11.md), Claude Code v2.1.321 (raw/anthropic-watch-claude-code-tag-v2-1-321.md).
- Claude Fable 5.2 and Mythos 5.2 launched 2026-11-09 at 40 per MTok (20% below 5.1’s 50) with $0.20 cache reads, output “up to 30% faster,” and about 55% fewer safeguard blocks than Fable 5.1.
- Claude Code moves you off 5.1. v2.1.321 made Fable 5.2 “the default Fable model” and “Changed saved Fable 5.1 model picks to move to fable : on Pro and Max plans at once, on Team and Enterprise plans once your organization allows it; use /model claude-fable-5-1 to switch back”.
- Plan terms (answers the AI Search question above). The Help Center says Fable 5.2, 5.1 and 5 “all work the same way”: included up to 50% of weekly limits on Max, Ultra and premium Team/Enterprise seats; pay-as-you-go usage credits on Pro and standard seats. So Fable 5.1 was available on Pro, but only through usage credits, not plan limits.
- Enterprise Frontier Safeguards moved to private beta with the 5.2 launch (Bridgewater, Salesforce, Stripe).
Migrating from Fable 5
From the claude-api skill’s “Migrating to Claude Fable 5.1 from Claude Fable 5” section (anthropics/skills #1704, 2026-09-01):
- Three breaking changes: (1) forced
tool_choiceofanyortoolreturns a 400; (2) thinking blocks are preserved only for the model that produced them or a newer one; (3) they are preserved only in the conversation that produced them, “so edited history replayed with thinking blocks is rejected.” - New API surface documented alongside: per-message effort (mid-conversation output-config beta), turn-scoped mid-conversation system messages with
clear_at, progress updates between tool calls viathinking.display: "updates", thinkingblock_bindingcontrols, and a 0.025× cache-read rate on Fable 5.1 with amax_tokens: 0keep-alive that “usually beats the 1-hour TTL.” - Launch-day hedges the skill kept open: Task Budgets support on Fable 5.1, whether Mythos 5.1 shares the 0.025× cache-read rate, and the fallback-credit wording.
- Claude Code v2.1.257 (2026-09-01):
claude-fable-5-1added and made the default Fable model (1M context, 50, $0.25 cache reads). In Claude apps gateway sessions,fableandbestkept resolving to Fable 5 for now, because gateways not yet configured for 5.1 reject it; pick Fable 5.1 in/modelexplicitly. See Week 36.
Implementation
Tool/Service: Claude Fable 5.1 (claude-fable-5-1); Claude Mythos 5.1 (claude-mythos-5-1, trusted access only)
Setup: Generally available on the Claude API, Amazon Web Services, Google Cloud, and Microsoft Azure. Claude Code v2.1.257+ uses it as the default Fable model except behind apps gateways (see above).
Cost: 50 output / $0.25 cache read per MTok.
Integration notes: Check any integration that forces tool_choice or replays edited history with thinking blocks before switching model IDs. Eligible enterprises should ask about zero data retention now rather than waiting for EFS.
Try It
- Re-price your Fable workloads. If your Fable bill is mostly cache reads (long agentic sessions), the ~45% estimate is the one that applies to you. Measure it on a week of real usage.
- Retry security work you stopped doing on Fable 5. Vulnerability discovery is now allowed, and interventions dropped ~60% per session.
- Pin subagents to a cheaper model. Put
model: sonnetormodel: opusin your subagent definitions, or say so in the prompt, so a Fable 5.1 coordinator doesn’t fan out Fable subagents. - Compare against Opus 5.5 before committing. Opus 5.5 (2026-09-22) claims Fable-5.1-level results on most work at 20. Keep Fable 5.1 for the tasks where it measurably wins.
Open Questions
- EFS timing and eligibility. “In phases, beginning later this fall”; which customers are “eligible” for interim ZDR is not defined on the launch page.
- Does ZDR end the adoption plateau? Anthropic’s Position, August 2026 recorded two competing explanations for Fable 5’s ~11% share of Anthropic tool spend: price or data retention. Anthropic naming both as customer feedback and fixing both at once means spend data after this release cannot separate them. ^[inferred]
- First spend data point (2026-09): Ramp data, relayed by The AI Daily Brief, had Fable 5.1 at 22.5% of enterprise spend “and rising very quickly” (
raw/The_AI_Challenges_Businesses_Are_Actually_Focused_On_Right_Now.md). That fits either explanation, since the release also cut running cost 25–45%.
- First spend data point (2026-09): Ramp data, relayed by The AI Daily Brief, had Fable 5.1 at 22.5% of enterprise spend “and rising very quickly” (
- Knowledge cutoff, max output, and tokenizer are not stated on the launch page; the system card (2026-09-01) was not read for this article.
- Fable 5.5. Rumoured from routing sightings and a reported 2026-10-13 Azure listing (see the October update); no release date or announcement in any source here.
- SDK model constant. No SDK release in this batch names
claude-fable-5-1; whether a constant landed is unconfirmed.
Related
- Claude Fable 5.2 + Mythos 5.2 — the successor (2026-11-09).
- Claude Fable 5 + Mythos 5 — the predecessor and its adoption problems.
- Claude Opus 5.5 — the cheaper model that claims parity on most work three weeks later.
- Claude Security — now powered by Mythos 5.1.
- Claude Output Watermarking — the EU AI Act watermark Fable 5.1 introduced.
- Anthropic’s Position, August 2026 — the retention-vs-price adoption question.
- Prompt Caching for Agencies — where a 75% cache-read cut lands.
- Week 36 digest — the Claude Code release that made it the default.
- Which Claude Model for Which Job — Anthropic’s own rule for when Fable 5.1 is worth it, with creator tests.
- Claude Tag — the Slack integration in the ops-review demo.