Capital & Compute

Claude Fable 5.1 Leak: Release Date and Price

Anthropic has published nothing on Fable 5.1. Four leaked claims, two X threads and one gray-scale sighting, graded against the primary sources.

· Updated August 25, 2026· ai· models· pricing· By Capital & Compute

Seven things are currently being asserted about Claude Fable 5.1. Six of them trace to a single X post. Zero of them appear in any document Anthropic has published. That ratio, checked on August 19, 2026, is the release information.

The latest addition is the interesting one. On August 19 the Chinese outlet 新智元, republished in English by 36Kr, reported that Fable 5.1 had leaked through a gray-scale test and that some people using the Claude web app or Claude Code may already have been served the new model without being told. The same report expects a full rollout before the end of August. It cites two X accounts. Nothing else corroborates it.

Every Fable 5.1 claim, and where it actually comes from

There are no leaked benchmarks, no leaked context window, no leaked model ID and no leaked system card. What exists is a short list of assertions, and it is worth seeing them separated from each other with their provenance attached, because the aggregation cycle has been repeating them as a set.

Claim Earliest named source Date Grade
A model called Fable 5.1 exists as a finished internal build X posts by community trackers Jul 26, 2026 Social
It ships in August 2026 @Mr_Salio on X, as reported by Times of AI Jul 26, 2026 Social
Pricing stays at $10 in and $50 out per million tokens LuminaXspace on X, relayed by 新智元 via 36Kr Jul 26 to 27, 2026 Social
The focus is long-horizon reasoning and agent work LuminaXspace on X Jul 26, 2026 Social
Anthropic is holding it back for OpenAI next release Andrew Curran on X, presented as his own read Jul 26, 2026 Opinion
A gray-scale rollout is live now inside claude.ai and Claude Code @synthwavedd and @kimmonismus, relayed by 新智元 via 36Kr Aug 19, 2026 Social
Full rollout before the end of August 新智元 projection, no named source Aug 19, 2026 Inference

Nothing in that table is from Anthropic. The Anthropic newsroom currently runs from a text-watermarking post on August 14 back through Improving Fable 5 biology safeguards on August 7 and Introducing Claude Opus 5 on July 24. Anthropic published a post specifically about Fable 5 twelve days ago and did not mention a successor. The platform documentation model list still names claude-fable-5 as the most capable widely released model, with no 5.1 row and no 5.1 alias.

Fable 5.1 claims by evidence grade, August 19 2026Seven claims about Claude Fable 5.1 are in circulation. Six trace to a named original source, in every case a post on X. Zero are confirmed by an Anthropic primary source: the newsroom, the platform documentation or the model list.-1 droppedClaims in circulationeverything asserted about Fable 5.17100% kept-6 droppedTraceable to a named sourcea dated X post or thread686% keptConfirmed by Anthropicnewsroom, docs or model list00% kept
Fable 5.1 claims by evidence grade, August 19 2026
StageTasks remainingPercent of candidate pool kept
Claims in circulation (everything asserted about Fable 5.1)7100%
Traceable to a named source (a dated X post or thread)686%
Confirmed by Anthropic (newsroom, docs or model list)00%
Claim attrition. Seven assertions are in circulation, six can be traced to a named original poster, and none is confirmed by a primary Anthropic source. Checked August 19, 2026.

That is not an argument that Fable 5.1 does not exist. Point releases are the most predictable thing Anthropic ships, and the cadence supports one: Fable 5 on June 9, Sonnet 5 on June 30, Opus 5 on July 24. It is an argument about what you can act on, which right now is nothing.

The chronology, official and unofficial

The leak reads differently once the official events are interleaved with it. Anthropic spent most of Fable 5’s first two months not shipping features but restoring access, and OpenAI repriced in the middle of that.

  1. June 9, 2026

    Official: Fable 5 and Mythos 5 ship

    General availability at $10 in and $50 out per million tokens, a 1M-token context window, 128k max output, and adaptive thinking always on.

  2. June 12, 2026

    Official: access suspended

    A US Commerce export-control directive led Anthropic to disable Fable 5 and Mythos 5 for every user, the first known use of that authority against a specific commercial frontier model.

  3. June 30, 2026

    Official: directive lifted

    Anthropic published Redeploying Fable 5 and began a staged restoration of general access in early July.

  4. July 9, 2026

    Official: GPT-5.6 goes general

    OpenAI released Sol, Terra and Luna publicly after a government-gated preview, with Sol at $5 in and $30 out.

  5. July 24, 2026

    Official: Claude Opus 5

    Anthropic shipped its flagship at $5 in and $25 out, half the Fable 5 rate, with an effort dial from low to max.

  6. July 25 to 27, 2026

    Unverified: the leak forms

    A WinCentral report, then X posts on July 26 claiming an August window, unchanged pricing and a long-horizon agent focus, then the 新智元 aggregation on July 27 describing Fable 5.1 as a held-back trump card.

  7. July 30, 2026

    Official: OpenAI cuts prices

    Terra fell 20% to $2 and $12, Luna fell 80% to $0.20 and $1.20, and flagship Sol was explicitly excluded and held at $5 and $30.

  8. August 1, 2026

    Official: OpenAI names Astra

    The next OpenAI frontier model got a name but no date, price, model card or availability, which leaves the wait-for-OpenAI theory untestable.

  9. August 7, 2026

    Official: Anthropic writes about Fable 5

    A newsroom post on Fable 5 biology safeguards, with no mention of a successor.

  10. August 19, 2026

    Unverified: the gray-scale report

    New reporting claims some claude.ai and Claude Code users are being served Fable 5.1 already, with a full rollout before month end. Sourced to two X accounts. The newsroom and the model list are still silent.

Fable 5 to Fable 5.1. Items marked official come from Anthropic or OpenAI publications; items marked unverified are third-party reports.

What a gray-scale rollout would and would not prove

A gray-scale, or canary, rollout is a real mechanism, and it is the most plausible part of the August 19 report. Vendors route a small share of traffic to a new build to gather production data before announcing. Anthropic does not label model swaps inside claude.ai the way it labels API model IDs, so a silent substitution is technically possible and users would be left inferring it from behavior.

The claim that matters is the price

Every aggregation of this leak leads with unchanged pricing as reassurance. Read as economics rather than as news, an unchanged $10 and $50 is the most demanding claim in the set, because the rest of the market moved underneath it.

0
Anthropic sources confirming Fable 5.1
newsroom, docs and model list, checked Aug 19 2026
$50
Output per million tokens, rumored to hold
the Fable 5 rate since June 9 2026
13.1 pts
GPT-5.6 Sol lead over Fable 5 on Agents' Last Exam
OpenAI-reported, Sol at 53.6
2x
Fable 5 price against Anthropic's own Opus 5
$10 and $50 versus $5 and $25

Three things happened between the Fable 5 launch and today. Anthropic shipped Opus 5 at $5 and $25, which Artificial Analysis rates at or above Fable 5 on its Intelligence Index at half the token price. OpenAI cut Terra by 20% and Luna by 80% on July 30, per the OpenAI API pricing documentation, then cut flagship Sol itself to $4 and $20 on August 21, 2026 on a promotional basis through at least November 21, 2026. And on Agents’ Last Exam, the 1,490-task benchmark of long-horizon professional work built with an advisory committee of more than 300 domain experts, OpenAI reported Sol setting a high of 53.6, ahead of Claude Fable 5 in adaptive mode by 13.1 points, and beating Fable 5 at medium reasoning by 11.4 points at roughly a quarter of the estimated cost. Those are OpenAI numbers on OpenAI runs, not independently reproduced. They are also on precisely the ground Fable 5.1 is rumored to target.

Frontier output prices per million tokens, August 25 2026Output price per million tokens on a log scale, as multiples of the $1.20 GPT-5.6 Luna floor: Claude Fable 5 at $50 is 42 times the floor, Claude Opus 5 at $25 is 21 times, GPT-5.6 Sol at its promotional $20 is about 17 times, GPT-5.6 Terra at $12 is 10 times, Claude Sonnet 5 at $10 is 8 times, Claude Haiku 4.5 at $5 is 4 times, and GPT-5.6 Luna is the $1.20 baseline. Fable 5 has the most expensive output rate on the chart, twice Claude Opus 5.$1.00$2.00$5.00$10.00$20.00$50.00output price, $ per million tokens (log scale)cheap floorClaude Fable 5$50.00 ×42Claude Opus 5$25.00 ×21GPT-5.6 Sol (promo)$20.00 ×17GPT-5.6 Terra$12.00 ×10Claude Sonnet 5$10.00 ×8Claude Haiku 4.5$5.00 ×4GPT-5.6 Luna$1.20 cheap floor
Frontier output prices per million tokens, August 25 2026
ToolCost per taskMultiple of baseline
Claude Fable 5$50.0041.7x
Claude Opus 5$25.0020.8x
GPT-5.6 Sol (promo)$20.0016.7x
GPT-5.6 Terra$12.0010.0x
Claude Sonnet 5$10.008.3x
Claude Haiku 4.5$5.004.2x
GPT-5.6 Luna$1.201.0x
Output rates today on a log scale, expressed as multiples of the cheapest frontier tier. Every rate is from vendor documentation; Sol is its promotional rate since August 21, 2026. The rumored Fable 5.1 rate is the $50 row: unchanged, and 42 times the floor.

So the price-hold claim, if true, means Anthropic is asking a 2x premium over its own flagship for a model whose distinguishing pitch is long-horizon agent work, in the same quarter that a competitor posted a 13-point lead on the leading long-horizon benchmark at a lower rate. That is a defensible position only if Fable 5.1 moves those numbers materially. It is why the useful question is not when it lands but what it scores, and the cost-per-task view is where the answer will show up first. On the current numbers, our own GPT-5.6 versus Fable 5 comparison already favors the cheaper model for most agent workloads.

There is a second reading worth stating, because it is the charitable one. Fable 5 carries safety classifiers that can refuse requests, 30-day data retention and no zero-retention option, and it spent its first three weeks unavailable under an export-control directive. A price hold in that context is not confidence, it is stability: changing the rate while the access story is still settling would add another variable for the enterprises that just rebuilt their fallback paths.

What would actually confirm Fable 5.1

Four artifacts, in the order they normally appear:

  1. A claude-fable-5-1 row in the platform model list, with pricing and a knowledge cutoff. This is the fact that ends the speculation.
  2. An Anthropic newsroom post. Every 2026 Claude release has had one, including Sonnet 5 and Opus 5.
  3. Bedrock, Vertex and Microsoft Foundry IDs, which usually land the same day.
  4. An OpenRouter listing with a rate card, which is where third-party pricing becomes checkable.

Until at least the first two exist, the entry on our AI model tracker stays at rumored with no price recorded, because writing $10 and $50 into a dataset on the strength of an X post is how a rumor becomes a fact. For what did actually ship this month, see new AI models released in August 2026.

Bottom line

The Fable 5.1 release information, in full, is an August window and an unchanged price, both from X posts on July 26, plus an uncorroborated report on August 19 that a canary is already live. Anthropic has said nothing, twelve days after publishing about Fable 5. Treat the August window as plausible on cadence alone, treat everything else as unconfirmed, and watch the price rather than the date: at $10 and $50 the interesting question is not whether Fable 5.1 arrives but whether it earns a 2x premium over Opus 5 on the agent work it is supposedly built for.

Frequently asked questions

Has Claude Fable 5.1 been released?
No. As of August 19, 2026 Anthropic has not announced Fable 5.1. There is no newsroom post, no claude-fable-5-1 model ID in the platform documentation, and no pricing. Reports of a gray-scale test serving the model to some claude.ai and Claude Code users come from two X accounts relayed by one outlet and are not corroborated.
When is the Claude Fable 5.1 release date?
There is no official date. Community trackers claimed on July 26, 2026 that an August 2026 release was planned, and an August 19 report expects a full rollout before the end of August. Anthropic's release cadence in 2026 makes an August or September window plausible: Fable 5 on June 9, Sonnet 5 on June 30, Opus 5 on July 24.
How much will Claude Fable 5.1 cost?
The only pricing claim in circulation is that it holds the Fable 5 rate of $10 per million input tokens and $50 per million output tokens. That figure comes from an X post on July 26, 2026, not from Anthropic, and no confirmed rate exists. For context, Claude Opus 5 costs $5 and $25 and GPT-5.6 Sol costs $4 and $20 on a promotional rate (August 21, 2026, through at least November 21, 2026).
Is Fable 5.1 better than Claude Opus 5?
Unknown. No benchmark for Fable 5.1 has leaked or been published. Today Fable 5 costs twice Opus 5 while Artificial Analysis rates Opus 5 at or above it on the Intelligence Index, so a 5.1 that keeps the $10 and $50 rate needs a clear win on long-horizon agent work to justify the premium.
What is a gray-scale rollout?
A gray-scale or canary rollout routes a small share of production traffic to a new build before it is announced, to gather real usage data. It shows a model exists and is being tested. It does not confirm the model name, its price, its specifications or its release date.

Sources

Get each breakdown before it makes the rounds

You get one email when a new source-backed analysis goes live: what AI agents actually cost, which models are worth running, and what the benchmarks really mean. No hype.

No spam. Unsubscribe anytime.

← Back to Models & benchmarks