Is claude-fable-5 back yet?

Yes — Fable 5 is back. FableWatch's monitor confirmed claude-fable-5 answering on July 1, 2026 at 7:29 PM ET and alerted the watch. Per Anthropic, it returned with a new safety classifier that routes some cybersecurity-adjacent requests to Opus 4.8. The live status at the top of this page tracks it continuously — and FableWatch now watches for the next Mythos-class model.

Why was claude-fable-5 suspended?

Anthropic suspended access to claude-fable-5 on June 12, 2026, reportedly after a US Commerce Department export-control directive — described in reporting as a letter from Secretary Howard Lutnick to Anthropic CEO Dario Amodei. The specifics are attributed to news outlets and remain partly unconfirmed; FableWatch is not affiliated with Anthropic and does not have inside information.

What is FableWatch watching now?

The next Mythos-class model. Fable 5 was the first Mythos-class release, and the next one — Fable 6, or a 5.1 — is the drop worth being first to. FableWatch pings the claude-fable-5 endpoint every 60 seconds and watches Anthropic's model lineup for new Mythos-class ids; the moment a new state-of-the-art model is live, subscribers are alerted. Our guarantee still stands: if you bought a paid tier and its alert never fired within 90 days of purchase, email us and we'll refund you in full. You're paying for the window, not a gamble.

Is this affiliated with Anthropic?

No. FableWatch isn't affiliated with, endorsed by, or connected to Anthropic. We built the independent monitor that caught claude-fable-5's July 1 return the minute it happened — and now we watch for the next drop.

Is it a subscription?

No. Every paid tier is a single one-time charge — The Signal and The Swarm both. Nothing recurring, nothing to cancel.

What exactly do I get for paying?

The Signal: you're alerted first — email and push fire instantly the minute a new state-of-the-art model answers, ahead of the free list, with SMS and an automated call to your verified number the same minute. The Swarm: all of that, plus a dedicated VPS that runs a 400+ question set in parallel on the new model on your own Anthropic key, plus GitHub fix suggestions — the Swarm drafts a suggested fix as a comment on your connected repos — with every answer appearing on your dashboard as it's captured (refresh to update).

When do SMS and calls actually fire?

Email and push are instant. SMS and an automated call are live too — US carrier registration (A2P 10DLC) has cleared, so once you verify your number, texts and calls fire the same minute Fable answers. And per our guarantee, if a paid tier never reaches you we refund it in full — you're never out money for a channel that couldn't deliver.

✳ the bench index

Every leaderboard, one table.

The public boards each tell part of the story — community votes here, live task suites there, lab-reported launch numbers somewhere else. This table reads them together: each model's best-variant score on every board we track, normalized to a percentile rank, and combined into one composite index. Refreshed daily, free, no signup.

Table last refreshed 2026-07-25· each board's own as-of date is listed with the sources below.

The index at a glancetrailingleading

Claude Opus 594.6

Claude Fable 594.1

GPT-5.688.2

Kimi K378.7

Claude Opus 4.875.7

GPT-5.563.4

Grok 4.557.9

Muse Spark 1.156.3

Claude Sonnet 554.4

Gemini 3.5 Flash41.2

GLM-5.240.6

Seed 2.1 Pro35.6

Gemini 3.1 Pro35.3

Qwen 3.7 Max30.3

MiniMax M317.6

DeepSeek V4 Pro17.6

Kimi K2.7 Code13.1

Kimi K2.612.5

#	Model	AW Index	LMArena	LiveBench	SimpleBench	LLM Stats	Vals AI	AA
1	Claude Opus 5Anthropic	94.6	—	—	80.6 93	58 96	74.8 94	60.7 100
2	Claude Fable 5Anthropic	94.1	1,635.8 94	80.8 94	81.9 100	57.5 82	75.1 100	59.9 94
3	GPT-5.6OpenAI	88.2	1,625 88	82.4 100	71.7 67	58 96	73.1 81	58.9 88
4	Kimi K3Moonshot	78.7	1,682 100	78.5 75	60.7 40	55.7 73	74.7 88	57.1 82
5	Claude Opus 4.8Anthropic	75.7	1,567.5 77	78.9 81	64.8 47	52.6 64	70.4 75	55.7 77
6	GPT-5.5OpenAI	63.4	1,525.3 47	79.9 88	76.9 80	49.1 27	68 56	54.8 71
7	Grok 4.5xAI	57.9	1,550.2 71	76.2 63	70 53	49.5 36	65.3 50	53.8 65
8	Muse Spark 1.1Meta	56.3	1,536.3 59	76.2 56	—	52.1 55	68.4 63	50.6 47
9	Claude Sonnet 5Anthropic	54.4	1,541.1 65	74.8 50	60.6 33	50.9 46	68.6 69	53.4 59
10	Gemini 3.5 FlashGoogle	41.2	1,492.6 24	74.6 44	76.7 73	—	62.7 38	50.2 41
11	GLM-5.2Zhipu AI	40.6	1,587.7 82	73.2 38	58.8 27	47.1 9	65 44	51.1 53
12	Seed 2.1 ProByteDance	35.6	1,528.3 53	—	—	47.3 18	—	—
13	Gemini 3.1 ProGoogle	35.3	1,485.8 12	77.1 69	79.6 87	—	53.8 6	46.5 35
14	Qwen 3.7 MaxAlibaba	30.3	1,518.7 41	73.1 31	70.4 60	46.7 0	57.5 25	46 29
15	MiniMax M3MiniMax	17.6	1,492.5 18	67.3 0	45.8 0	—	58.9 31	44.4 24
16	DeepSeek V4 ProDeepSeek	17.6	1,464.2 6	71.6 19	50.9 13	—	55.6 19	44.3 18
17	Kimi K2.7 CodeMoonshot	13.1	1,517.4 29	68.4 6	57.9 20	—	—	41.9 6
18	Kimi K2.6Moonshot	12.5	1,518.7 35	70.5 13	—	—	55.2 13	44.2 12
19	InklingThinking Machines	0	1,444.9 0	71.7 25	50 7	—	49.3 0	40.7 0

Cell format: raw board score + percentile chip. Hover any cell for the exact variant behind the number.

How the index is computed

01Best variant wins

Labs ship many configs — thinking modes, effort levels, harnesses. Per board, each model is scored by its best-performing variant, identically for everyone. Search, grounding, image and video product modes are excluded.

02Percentiles, not raw averages

Elo ≈1500 and 0–100 scores can't be averaged. Each board's scores become percentile ranks (0–100) within the set of models on this table that the board actually scores.

03The index is a median

A model's AW Index is the median of its percentiles across boards — so no single board's methodology can dominate. A model needs 2+ boards to be ranked; boards scoring fewer than 5 of these models display but don't vote.

04Nothing is estimated

Every raw number comes from the source board, links back to it, and carries its fetch date. Percentiles are relative to this table's model set, not the boards' full lists.

The boards

LMArena ↗community votes — millions of blind head-to-headsElo (community votes) · 18 models mapped · as of 2026-07-25 LiveBench ↗independent — contamination-free live task suiteglobal average (0-100) · 17 models mapped · as of 2026-07-21 SimpleBench ↗independent — private reasoning set, human-baselinedAVG@5 % (private set) · 16 models mapped · as of 2026-07-25 LLM Stats ↗LLM Stats' headline composite — verified benchmarks + live perfLLM Stats Score · composite (0-100), their rendered top 15 · 12 models mapped · as of 2026-07-25 Vals AI ↗independent — expert-crafted private tasks, measured runsVals Index accuracy % · weighted finance + coding tasks · 17 models mapped · as of 2026-07-25 Artificial Analysis ↗independent — standardized eval suite, measured runs onlyintelligence index · measured runs only · 18 models mapped · as of 2026-07-25

Benchmarks tell you who leads today; the radar tells you the minute that changes. Watch the models · the wire · track record