Claude Fable 5
Claude Fable 5 was the first Mythos-class model Anthropic opened to all customers. It shipped on Jun 9, 2026, went offline from Jun 12 to Jul 1 under a US government order and is now legacy, replaced by Fable 5.1 on Sep 1, 2026. The successor charges the same US$ 10 per million input tokens and US$ 50 per million output, but rereads the cache for US$ 0.25 instead of US$ 1, and Anthropic estimates the bill about 25% lower on typical workloads. Retirement will not come before Jun 9, 2027; it is only worth staying if you rely on forced tool use or Priority Tier, which 5.1 does not support.
Claude Fable 5 in official numbers
Data published by the maker in its documentation and announcement. The source of each item is at the end of the page.
- Maker
- Anthropic
- API identifier
claude-fable-5- Release
- Jun 9, 2026
- Status
- legacy; the successor is Fable 5.1, released on Sep 1, 2026
- Input
- US$ 10 per million tokens
- Output
- US$ 50 per million tokens
- Cache read
- US$ 1 per million (10% of input; US$ 0.25 on Fable 5.1)
- Cache write
- US$ 12.50 (5 minutes) or US$ 20 (1 hour)
- Batch API
- US$ 5 input and US$ 25 output
- Fast mode
- not available for Fable 5
- Context window
- 1 million tokens, no surcharge
- Max output
- 128 thousand tokens
- Reliable knowledge
- through January 2026
- Reasoning
- adaptive, always on; default effort high; no mid-conversation effort changes
- Where it runs
- Anthropic API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS
- Data retention
- 30 days; zero retention only with Anthropic's authorization
- Retirement
- not before Jun 9, 2027
Where Claude Fable 5 does well, where it falls behind and who it is for
Where it does well
- Agentic coding: 80% on SWE-bench Pro and 95% on SWE-bench Verified in the system card, against 69.2% and 88.6% for Opus 4.8.
- Office work: 1,932 Elo on GDPval-AA, against 1,890 for Opus 4.8 and 1,769 for GPT-5.5, in the system card.
- Vision and documents: 29.8% on GDP.pdf without tools and 38.6% on Blueprint-Bench 2, against 22.5% and 14.5% for Opus 4.8.
- Hard coding: 29.3% on FrontierCode Diamond, against 13.4% for Opus 4.8 and 5.7% for GPT-5.5.
- Accepts what 5.1 removed: it still allows forced tool use with
tool_choiceand runs on Priority Tier.
Where it falls behind
- Expensive cache: cache reads cost US$ 1 per million; Fable 5.1 charges a quarter of that, US$ 0.25.
- Overtaken by its successor: in the Fable 5.1 table, it trails on all nine rows, such as Terminal-Bench 4.0 (42.0% against 55.8%) and Terminal-Bench-Science (24.7% against 52.6%).
- Behind Opus 5, which costs half: at the Opus 5 launch, it lost on Frontier-Bench v0.1 (33.7% against 43.3%) and GDPval-AA v2 (1,747 against 1,861).
- Broad safeguards: cyber, biology, chemistry and distillation requests are rerouted to Opus models, and the July classifier flags more harmless coding requests.
- Retention required: data is kept for 30 days, and the API rejects organizations without that retention with a 400 error.
Who it makes sense for
- Integrations that force a tool with
tool_choiceanyortool: on Fable 5.1, the same call returns a 400 error. - Organizations with a Priority Tier contract: Fable 5 supports it, Fable 5.1 does not.
- Code that rewrites history between turns: on Fable 5.1, accounts created on or after Aug 31, 2026 get a 400 error in that case.
- Systems validated on Fable 5 that need time to rerun their tests, with the guarantee it will not be retired before Jun 9, 2027.
When to pick another model
- New projects: Fable 5.1 costs the same per token, charges a quarter for cache reads and is what the documentation points to.
- As a starting point: Anthropic says to start with Opus 5.5, at US$ 4 and US$ 20, and move up to Fable 5.1 only if it falls short.
- Apps with heavy caching, such as a large repository: that is where 5.1 cuts the bill most, up to about 45% on highly agentic workloads, according to Anthropic.
- Quick answers or live chat: there is no fast mode, and a turn at high effort can take minutes.
What Anthropic published at launch
Fable 5 shipped on Jun 9, 2026 together with Mythos 5, the same model without safeguards, limited to Project Glasswing. On Fable 5, classifiers hand cybersecurity, biology and chemistry, and distillation requests to Opus 4.8; at launch, Anthropic said this happened in less than 5% of sessions.
The announcement table shows the higher score of Mythos 5 and Fable 5, and on starred tests the number belongs to Mythos. That is why the table below uses Fable 5's own column in the system card (max effort, averaged over 5 trials).
| Fable 5 Anthropic | Mythos 5 Anthropic | Opus 4.8 Anthropic | GPT-5.5 OpenAI | |
|---|---|---|---|---|
| SWE-bench Pro coding |
80 | 80.3 | 69.2 | 58.6 |
| SWE-bench Verified coding |
95 | 95.5 | 88.6 | not published |
| Terminal-Bench 2.1 terminal coding |
84.3 | 88.0 | 82.7 | 83.4 |
| OSWorld-Verified computer use |
85.0 | 85.0 | 83.4 | 78.7 |
| FrontierCode Diamond hard coding |
29.3 | not published | 13.4 | 5.7 |
| GDPval-AA office work, Elo |
1,932 | not published | 1,890 | 1,769 |
| GDP.pdf document reading |
29.8 | not published | 22.5 | 24.9 |
| OfficeQA Pro office documents |
57.9 | not published | 48.1 | 52.6 |
| AutomationBench business workflows |
17.4 | not published | 15.5 | 12.9 |
| Blueprint-Bench 2 spatial reasoning |
38.6 | not published | 14.5 | 36.2 |
| Legal Agent Benchmark legal, Harvey set |
13.3 | not published | 10.4 | 2.1 |
Claude Fable 5 and Claude Mythos 5 system card, table 8.1.A. Fable 5's scores already include the fallback to Opus 4.8 when the safeguards act. GDPval-AA is the Elo as of Jun 6, 2026. The dot marks the best in each row.
The June suspension
Three days after launch, the US government applied export controls to Fable 5 following a report by Amazon researchers on a way around the safeguards. With no way to check users' nationality, Anthropic switched the model off for everyone. It came back with a classifier that blocks the technique in over 99% of cases, at the cost, as the company admitted, of blocking more harmless coding requests.
- Jun 9, 2026: launch; on subscription plans, included at no extra cost through Jun 22.
- Jun 12, 2026: US government directive and suspension for everyone.
- Jun 30, 2026: the export controls are lifted.
- Jul 1, 2026: Fable 5 returns on the Claude Platform, Claude.ai, Claude Code and Cowork; on Pro, Max, Team and some Enterprise plans, included for up to 50% of weekly limits through Jul 7, then only with usage credits.
Is Fable 5 still worth using in 2026?
In most cases, no. Anthropic treats Fable 5 as legacy and recommends Fable 5.1, which charges the same for input and output and a quarter for cache reads. In Anthropic's chart, based on real usage in August 2026, the bill falls about 25% on typical workloads and 45% on highly agentic ones, and the company attributes the savings to cheaper cache reads.
Two reasons to wait remain: Fable 5.1 does not accept forced tool use and does not run on Priority Tier. If you depend on either, you have until retirement, which will not come before Jun 9, 2027, with at least 60 days' notice. And there is Opus 5.5: at US$ 4 and US$ 20, it beat Fable 5.1 on all nine tests in its own launch.
| Fable 5 legacy | Fable 5.1 current | Opus 5.5 current | |
|---|---|---|---|
| Input | US$ 10 | US$ 10 | US$ 4 |
| Output | US$ 50 | US$ 50 | US$ 20 |
| Cache read | US$ 1 | US$ 0.25 | US$ 0.20 |
| Cache write 5 minutes |
US$ 12.50 | US$ 12.50 | US$ 5 |
| Batch input and output |
US$ 5 and US$ 25 | US$ 5 and US$ 25 | US$ 2 and US$ 10 |
| Default effort | high | high | medium |
| Reliable knowledge | January 2026 | June 2026 | June 2026 |
| Retirement not before |
Jun 9, 2027 | Sep 1, 2027 | Sep 22, 2027 |
Anthropic documentation, pricing, models overview and model deprecations, checked on Oct 1, 2026.
What changes when you migrate to Fable 5.1
The API, limits, per-token price, tokenizer, always-on thinking and refusal categories stay the same. What changes, according to Anthropic's guide:
- Forced tool use:
tool_choiceanyandtoolreturn a 400 error. Useautowith an explicit instruction and strict tools. - Thinking tied to the conversation: Fable 5.1's blocks are valid only for it (or a newer model) and only in the history where they were produced. On accounts created on or after Aug 31, 2026, resending a block after changing the system prompt, the tools or an earlier message returns a 400 error.
- Fewer parallel calls: in long loops, 5.1 may make one call per turn; Anthropic suggests a one-sentence instruction asking it to batch calls.
- Fewer progress messages and fewer searches at low effort; in exchange, effort can change mid-conversation (beta) without losing the cache.
Examples of using Claude Fable 5
A pipeline reads invoices and forces the model to answer through the record tool. On Fable 5 it works; on Fable 5.1 it returns a 400 error. Anthropic's suggested fix is auto with an explicit instruction and strict tools.
The whole codebase goes into the context and stays cached. On Fable 5, each reread costs US$ 1 per million tokens; on Fable 5.1, US$ 0.25. This is the kind of usage where Anthropic measured the largest cost drop.
The system runs the task on Opus 5.5 and only moves up to a Fable model when verification fails. It is the rule in Anthropic's documentation, and it avoids paying US$ 50 per million output tokens for what Opus solves at US$ 20.
Where Claude Fable 5 stands out and where it falls behind, by the numbers
Claude Fable 5 has published results on 37 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Sep 12, 2025 and Sep 7, 2026 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.
Where it stands out
Tests where it sits in the top third of its own generation.
Where it falls behind
Tests where it sits in the bottom half of its own generation.
See all 37 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| GPQA Diamond | 85.9% | #51 / 74 (tied with 1) | #64 / 173 | 96.4% Gemini 4 Argon |
| AIME 2024 and 2025 (OTIS) | 100.0% | #1 / 76 (tied with 6) | #1 / 168 | 100.0% GPT-5.5 |
| Dtbench | 98.4% | #1 / 71 | #2 / 148 | 98.9% Claude Opus 5.5 |
| Lmca | 61.15 | #4 / 69 | #5 / 121 | 68.23 Claude Opus 5.5 |
| CritPt | 28.6% | #8 / 69 | #12 / 112 | 32.3% GPT-5.6 Sol |
| WeirdML | 91.9% | #3 / 60 | #3 / 112 | 93.6% GPT-6 Astra |
| SciCode | 61.0% | #2 / 65 | #4 / 104 | 66.9% Claude Opus 5.5 |
| Frontiermath Tiers 1 3 V2 | 87.0% | #5 / 59 | #9 / 86 | 93.7% GPT-6 Astra |
| ARC-AGI 2 | 89.2% | #5 / 49 (tied with 1) | #8 / 83 | 95.0% GPT-6 Astra |
| ALE-Bench | 2,041 | #4 / 60 | #7 / 82 | 2,951 GPT-6 Astra |
| WebDev Arena | 1,628 | #7 / 69 | #13 / 82 | 1,820 Claude Opus 5.5 |
| ARC-AGI | 98.5% | #1 / 49 (tied with 2) | #1 / 81 | 98.5% Claude Fable 5 |
| SimpleQA Verified | 70.7% | #4 / 56 | #7 / 81 | 76.0% Gemini 4 Argon |
| Chess puzzles | 47.0% | #15 / 57 (tied with 1) | #22 / 80 | 76.0% GPT-6 Astra |
| Game puzzles | 52.0% | #6 / 53 | #11 / 77 | 84.0% GPT-6 Astra |
| ProofBench | 95.0% | #4 / 56 | #8 / 74 | 100.0% Claude Fable 5.1 |
| SimpleBench | 81.9% | #1 / 41 | #1 / 74 | 81.9% Claude Fable 5 |
| Frontiermath Tier 4 V2 | 90.2% | #2 / 48 | #4 / 67 | 100.0% GPT-6.1 Sol |
| Vending Bench 2 | 5,680 | #18 / 47 | #23 / 57 | 15,515 GPT-6 Astra |
| Apex Agents | 63.6% | #7 / 39 | #10 / 50 | 82.4% Gemini 4 Argon |
| FrontierCode | 53.5% | #1 / 32 | #2 / 43 | 54.6% Claude Opus 5.5 |
| Gdp Pdf | 30.0% | #3 / 28 | #7 / 37 | 35.0% GPT-6.1 Sol |
| EnigmaEval | 39.3% | #1 / 17 | #1 / 36 | 39.3% Claude Fable 5 |
| Furniture Assembly | 35.8% | #12 / 23 | #20 / 34 | 83.3% Claude Opus 5.5 |
| Blueprint Bench 2 | 38.6% | #3 / 25 (tied with 1) | #5 / 31 | 54.4% Gemini 4 Argon |
| Gso | 78.4% | #3 / 18 | #3 / 28 | 88.2% Claude Fable 5.1 |
| Ebr Bench | 39.5% | #5 / 19 | #10 / 27 | 76.2% GPT-6 Astra |
| DeepSWE | 69.9% | #5 / 26 | #5 / 26 | 74.1% GPT-6 Astra |
| Surface Evolver Bench | 95.0% | #1 / 21 (tied with 1) | #1 / 24 | 95.0% Claude Fable 5 |
| Gbaeval | 74.5% | #2 / 22 | #2 / 22 | 79.6% Claude Opus 5 |
| FrontierSWE | 47.0% | #4 / 15 | #7 / 20 | 65.5% GPT-6 Astra |
| Rli | 16.1% | #3 / 12 | #4 / 15 | 21.3% Claude Opus 5.5 |
| Posttrainbench | 41.8% | #1 / 11 | #1 / 11 | 41.8% Claude Fable 5 |
| Weirdml V3 | 15.6% | #3 / 7 | #5 / 11 | 42.2% GPT-6 Astra |
| MirrorCode | 63.9% | #2 / 8 | #3 / 9 | 77.4% Claude Opus 5.5 |
| Btf3 | 13.0% | #8 / 8 | #8 / 8 | 14.9% GLM-5.3 |
| Frontiermath Erdos | 0.0% | #2 / 5 (tied with 3) | #4 / 7 | 2.9% GPT-6 Astra |
For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.
What to compare Claude Fable 5 with
The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.
| Model | Lab | ECI | ELO | LiveBench | Release |
|---|---|---|---|---|---|
| Gemini 4 Argon | Google DeepMind | 164.56 | 1,608 | n/a | Sep 30, 2026 |
| Claude Fable 5 | Anthropic | 162.13 | 1,542 | 83.0 | Jun 9, 2026 |
| GPT-5.5 Pro | OpenAI | 161.97 | n/a | n/a | Apr 23, 2026 |
| GPT-5.6 Sol | OpenAI | 161.73 | 1,527 | 81.1 | Jul 9, 2026 |
| Kimi K3 | Moonshot | 157.42 | 1,528 | 79.2 | Jul 16, 2026 |
| Gemini 3.7 Flash | Google DeepMind | 157.22 | 1,551 | 78.8 | Aug 13, 2026 |
| Qwen 3.8 Max | Alibaba | 156.88 | 1,536 | 78.5 | Aug 2, 2026 |
Claude Fable 5 or GPT-5.5 Pro?
GPT-5.5 Pro is the model from another lab closest to Claude Fable 5 in capability. Across the 11 tests both took, this is the score:
| Test | Claude Fable 5 | GPT-5.5 Pro |
|---|---|---|
| Chess puzzles | 47.0% | 62.0% |
| Frontiermath Tier 4 V2 | 90.2% | 78.0% |
| Lmca | 61.15 | 53.94 |
| GPQA Diamond | 85.9% | 93.9% |
| CritPt | 28.6% | 30.6% |
| SimpleBench | 81.9% | 76.9% |
| ARC-AGI 2 | 89.2% | 84.6% |
| Dtbench | 98.4% | 96.0% |
| ARC-AGI | 98.5% | 96.5% |
| Frontiermath Tiers 1 3 V2 | 87.0% | 87.7% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
What Claude Fable 5 costs to deliver each task
Price per token is what is on the price sheet; cost per task is what lands on the invoice, because it depends on how many tokens the model spends. Each point is an effort level, published by whoever ran the test and compiled by Epoch AI. The further up and to the left, the more score per dollar.
DeepSWE
The score scale does not start at zero.
Other models from Anthropic
The models from the same lab closest in time, newest first.
| Model | Release | ECI | ELO | LiveBench | Access |
|---|---|---|---|---|---|
| Claude Sonnet 5.5 | Sep 28, 2026 | 164.45 | 1,537 | 77.8 | API access |
| Claude Opus 5.5 | Sep 22, 2026 | 166.82 | 1,584 | 83.2 | API access |
| Claude Fable 5.1 | Sep 1, 2026 | 164.65 | 1,586 | 83.4 | API access |
| Claude Opus 5 | Jul 24, 2026 | 162.60 | 1,573 | 80.1 | API access |
| Claude Sonnet 5 | Jun 30, 2026 | 156.36 | 1,491 | 76.0 | API access |
| Claude Fable 5 | Jun 9, 2026 | 162.13 | 1,542 | 83.0 | API access |
| Claude Opus 4.8 | May 28, 2026 | 158.26 | 1,507 | 76.2 | API access |
| Claude Opus 4.7 | Apr 16, 2026 | 156.32 | 1,532 | 76.5 | API access |
| Claude Sonnet 4.6 | Feb 17, 2026 | 152.39 | 1,504 | 73.0 | API access |
Have you used Claude Fable 5? Rate it
Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.
What people ask about Claude Fable 5
Is Claude Fable 5 still available?
Yes, as a legacy model, on the Anthropic API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Anthropic guarantees it will not be retired before June 9, 2027 on the platforms it operates and gives at least 60 days' notice.
Why did Fable 5 go offline in June 2026?
On June 12, 2026, the US government applied export controls to Fable 5 and Mythos 5, after an Amazon report on a way around the model's safeguards. With no way to verify users' nationality, Anthropic switched both off for everyone. The controls were lifted on June 30, and Fable 5 returned on July 1 with a new classifier.
How much does Claude Fable 5 cost?
US$ 10 per million input tokens and US$ 50 per million output. Cache reads cost US$ 1, cache writes US$ 12.50 (5 minutes) or US$ 20 (1 hour), and batch is US$ 5 and US$ 25. There is no fast mode. Since September 24, 2026, refusals before any output are billed in the bio, frontier_llm and reasoning_extraction categories.
What is the difference between Fable 5 and Mythos 5?
They are the same model, with the same price and specifications. Fable 5 has classifiers that refuse or reroute cybersecurity, biology and chemistry, and distillation requests; Mythos 5 does not and is offered by invitation only, through Project Glasswing.
Where does Claude Fable 5 rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, Claude Fable 5 ranks #9 out of 267 models, with an ECI of 162.13. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.
Is Claude Fable 5 better than GPT-5.5 Pro?
Across the 11 tests both models took and Epoch AI compiled, Claude Fable 5 comes out ahead on 6, behind on 4 and tied on 1. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.
What is Claude Fable 5 best at?
Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of Claude Fable 5 are: Dtbench (#1 of 71); SimpleBench (#1 of 41); FrontierCode (#1 of 32). The yardstick is its position among the models of that period that took the same test.
Where do the numbers on this page come from?
From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.
Where each number comes from
- Anthropic, Claude Fable 5 and Claude Mythos 5 announcement
- Anthropic, statement on the Fable 5 suspension (Jun 12, 2026)
- Anthropic, redeploying Fable 5 (Jun 30, 2026)
- Anthropic, Claude Fable 5 and Claude Mythos 5 system card
- Anthropic, Claude Fable 5 overview (documentation)
- Anthropic, introducing Claude Fable 5 and Claude Mythos 5 (documentation)
- Anthropic, migrating to Claude Fable 5.1 (documentation)
- Anthropic, pricing (documentation)
- Anthropic, model deprecations (documentation)
- Anthropic, platform release notes (documentation)
- Anthropic, Claude Fable 5.1 announcement (comparison and indexed cost)
- Anthropic, Claude Opus 5 announcement (comparison with Fable 5)
Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.
The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.
LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.
← See the full ranking