Evidence checked November 9, 2026, the day Anthropic released Claude Fable 5.2. Prices are direct Claude API rates in US dollars. Kingy has not tested Fable 5.2; this page uses published evidence.
Claude Opus 5.5 is still the better starting choice for a new Claude coding or knowledge-work workflow. Anthropic’s own launch post for Fable 5.2 recommends starting most coding and knowledge work on Opus 5.5 and moving up to Fable when the work needs it. What Fable 5.2 changes is the size of the step up. It costs 20% less than Fable 5.1, so its ordinary input and output prices are now 2 times Opus’s and no longer 2.5 times. It also tops Cursor’s coding table and most rows of Anthropic’s launch table. Anthropic’s Fable 5.2 announcement.
Cursor’s CursorBench 4.0 shows the trade. At max effort, Fable 5.2 scores 60.0% at $32.79 per task, the top row of the table. Opus 5.5 max scores 57.8% at $13.43. That is 2.2 more points for $32.79 ÷ $13.43 = 2.44 times the cost. At high effort the gap is 0.9 points (56.9% against 56.0%) and the cost ratio is again 2.44 ($9.70 against $3.97). Cursor’s evaluation table.
Small score gaps like these may not be meaningful on their own. Whether they repay the premium depends on effort, cached context and the kind of work, so the sections below take each in turn.
What changes between the models
| Specification or rate | Claude Opus 5.5 | Claude Fable 5.2 | Claude Fable 5.1 (legacy) |
|---|---|---|---|
| Direct API ID | claude-opus-5-5 |
claude-fable-5-2 |
claude-fable-5-1 |
| Released | September 22, 2026 | November 9, 2026 | September 1, 2026 |
| Context window | 1M tokens | 1M tokens | 1M tokens |
| Standard maximum output | 128K tokens | 128K tokens | 128K tokens |
| Modalities | Text/images in; text out | Text/images in; text out | Text/images in; text out |
| Adaptive thinking | Always on | Always on | Always on |
| Default effort | medium | high | high |
| Ordinary input per 1M tokens | $4 | $8 | $10 |
| Output per 1M tokens | $20 | $40 | $50 |
| Cache read per 1M tokens | $0.20 | $0.20 | $0.25 |
| Five-minute cache write per 1M tokens | $5 | $10 | $12.50 |
| One-hour cache write per 1M tokens | $8 | $16 | $20 |
Sources: Opus 5.5 model reference, Fable 5.2 model reference, Fable 5.1 model reference. Anthropic now labels Fable 5.1 a legacy model and suggests moving to Fable 5.2.
All three support low, medium, high, xhigh and max effort. These are behavioral controls rather than fixed token budgets. The model can change its reasoning, tool use and output at each level. Leaving effort unspecified gives you medium on Opus and high on Fable, so set it explicitly when comparing them. Anthropic’s effort guide.
The cache row has changed the most. Fable 5.2’s cache-read rate is $0.20 per million tokens, the same as Opus’s. Against Fable 5.1 Opus still had a small edge there ($0.20 against $0.25). A session dominated by cache reads now costs the same on either model for those tokens, and the saving from choosing Opus comes only from fresh input, cache writes and output.
Coding at every effort setting
CursorBench 4.0 reports these scores and average costs. Each cell is score / cost per task.
| Effort | Opus 5.5 | Fable 5.2 | Fable 5.1 |
|---|---|---|---|
| Low | 43.7% / $1.17 | 53.2% / $3.94 | 45.1% / $5.44 |
| Medium | 52.5% / $2.91 | 54.8% / $4.66 | 46.8% / $7.05 |
| High | 56.0% / $3.97 | 56.9% / $9.70 | 49.2% / $9.08 |
| Xhigh | 56.0% / $6.98 | 56.3% / $15.48 | 51.6% / $13.01 |
| Max | 57.8% / $13.43 | 60.0% / $32.79 | 51.8% / $17.28 |
Source: CursorBench 4.0, read November 9, 2026. Cursor labels the fourth level “Extra High”. Cursor evaluates ambiguous, multi-file coding tasks and prices the tokens consumed at published rates. Small score differences may not be statistically meaningful. Methodology.
Fable 5.2 has the higher displayed score at every matching label, and the higher cost at every one. The widest gap is at low effort: 53.2 − 43.7 = 9.5 points, for $3.94 ÷ $1.17 = 3.37 times the cost. From medium to xhigh the gaps are 2.3, 0.9 and 0.3 points.
Matching labels is not the fairest comparison, because the two models cost very different amounts at the same label. Compare by cost instead. Opus high scores 56.0% at $3.97. Fable 5.2 low costs almost the same, $3.94, and scores 53.2%, which is 2.8 points lower. Fable 5.2 medium scores 54.8% at $4.66: Opus high is 1.2 points higher and ($4.66 − $3.97) ÷ $4.66 = 14.8% cheaper. On this table, Opus high is the configuration to beat below about $5 a task.
Above that the picture is closer. Fable 5.2 high is 0.9 points below Opus max at $9.70 ÷ $13.43 = 72.2% of the cost. Only Fable 5.2 max scores higher than Opus max, and it adds 3.1 points over its own high setting for 3.38 times the bill. Its xhigh row is 0.6 points below its high row while costing 59.6% more, which is a reminder that a higher setting does not guarantee a higher score.
The Opus advice from September still holds. Moving Opus from medium to high adds 3.5 points for 36.4% more cost. Moving from high to xhigh changes the displayed score by zero while cost rises 75.8%. Max costs 3.38 times the high-effort bill.
Against its predecessor, Fable 5.2 is a clear step. At low effort it scores 8.1 points higher than Fable 5.1 and costs 27.6% less per task. At max it scores 8.2 points higher but costs 1.90 times as much: Cursor’s table shows it using about 3 times the tokens per task at that setting.
What Anthropic’s launch numbers show
Anthropic’s announcement compares the three models at each one’s highest available effort unless a row says otherwise. These are the scores Anthropic published, with safeguards on.
| Evaluation | Fable 5.2 | Opus 5.5 | Fable 5.1 |
|---|---|---|---|
| Agentic coding: FrontierCode 2.0 | 49.1% (xhigh) | 45.5% | 32.4% (xhigh) |
| Agentic coding: FrontierSWE v2 | 69.9% | 62.3% | 56.3% |
| Business workflows: AutomationBench | 51.4% | 42.5% | 31.4% |
| Agentic search: WANDR (relaxed score) | 82.1% | 72.3% | 68.7% |
| Computer use: OSWorld 2.1 (offline set, partial) | 89.3% | 87.2% | 83.5% |
| Knowledge work: AA-Briefcase v1.1, rubric | 66.1% | 61.2% | 61.5% |
| Knowledge work: AA-Briefcase v1.1, Elo | 1785 | 1800 | 1669 |
| Knowledge work: Remote Labor Index | 41.3% (high) | 21.3% | 17.9% |
| Reasoning: Humanity’s Last Exam (with tools) | 74.2% | 67.7% | 65.6% |
| Chart reading: Chartography (no tools) | 73.4% | 64.4% | 44.8% |
Source: Anthropic’s Fable 5.2 benchmark table, read November 9, 2026. The table also lists cybersecurity, protein-design, science and robotics rows that are not repeated here.
Fable 5.2 leads Opus 5.5 on every row above except one: on AA-Briefcase, Fable 5.2 has the higher rubric score (66.1% against 61.2%) and Opus has the higher Elo rating (1800 against 1785). The two measures disagree, so that benchmark does not settle the knowledge-work question either way.
Anthropic also makes efficiency claims that cut against the simple “Fable costs twice as much” reading. On FrontierCode 2.0 it reports Fable 5.2 at high effort scoring 45.3%, about level with Opus 5.5’s best score of 45.5%, for roughly half the cost per task. On WANDR it reports Fable 5.2 at medium matching Opus’s top score for about 40% of the cost. On Chartography it reports Fable 5.2 at medium scoring 66.4%, above Opus at max (64.4%), for less than a tenth of the cost. Anthropic’s charts do not print a dollar figure for each point, so treat these as reasons to test, not as a price list.
Cursor’s independent table agrees with the direction and not the size. There, Fable 5.2 high sits just under Opus max at 72.2% of the cost, which is a saving but not half.
Kingy’s own November 9 test of 11 models included Opus 5.5 and Fable 5.1 but not Fable 5.2. See those results and costs.
An independent check: Artificial Analysis
Artificial Analysis runs all three models through the same index of nine evaluations. Each column is the model at max effort with Anthropic’s default fallback.
| Intelligence Index v5.0.1 | Opus 5.5 | Fable 5.2 | Fable 5.1 |
|---|---|---|---|
| Composite score | 59 | 65 | 49 |
| Rank among 74 listed models | 5 | 1 | 24 |
| Estimated cost per index task | $10.04 | $26.54 | $12.99 |
| Time per index task | 25.1 minutes | 46.4 minutes | 22.5 minutes |
Source: Artificial Analysis model pages for Opus 5.5, Fable 5.2 and Fable 5.1, read November 9, 2026. The time figures are live measurements and move slightly between readings.
- Fable 5.2 buys 6 more index points for about 2.6 times the cost ($26.54 ÷ $10.04) and about 1.8 times the wait (46.4 ÷ 25.1 minutes). Its token prices are only 2 times Opus’s; the rest of the gap is Fable using more tokens per task.
- Opus 5.5 beat Fable 5.1 outright: 10 points higher, and cheaper per task. Among these three, only Fable 5.2 scores above Opus.
- These are index points, not percentages, and an index task is a weighted benchmark unit, not a finished job. Default fallback means some answers may have been served by a different model after a refusal.
This matches the Cursor result above: a small lead for Fable 5.2 at well over double the cost. It supports the same advice. Start on Opus 5.5 and move to Fable 5.2 for the work where the extra points matter.
The September comparison: Opus 5.5 vs Fable 5.1
This section is kept from the September 22, 2026 version of this page. It compares Opus 5.5 with Fable 5.1, the Fable model sold at the time, using charts Anthropic published when it launched Opus 5.5. Anthropic has not published the same two charts for Fable 5.2, so these numbers describe the older model only.
Anthropic’s own Terminal-Bench 4.0 curve provides a second view:
| Effort | Opus 5.5 score / cost per attempt | Fable 5.1 score / cost per attempt |
|---|---|---|
| Low | 38.5% / $1.29 | 40.2% / $5.70 |
| Medium | 57.6% / $2.94 | 43.4% / $7.80 |
| High | 64.2% / $3.88 | 49.4% / $10.50 |
| Xhigh | 66.4% / $7.35 | 51.3% / $15.80 |
| Max | 64.8% / $11.24 | 55.8% / $19.50 |
Source: Anthropic’s coding charts.
Opus medium is 1.8 points above Fable max at $2.94 ÷ $19.50 = 15.1% of the displayed cost. The small score gap should not be sold as a conclusive capability win: Anthropic reports a ±2.6-point standard error for its headline Opus result and roughly ±1.6–2 points for the other Claude models. The cost gap is a stronger reason to investigate the cheaper configuration. Benchmark footnotes.
Knowledge work gave Fable 5.1 a more nuanced comparison
Anthropic’s GDPval-AA v2.1 chart measures professional work using Elo ratings, not task-completion percentages. Here are its displayed ratings and estimated costs:
| Effort | Opus 5.5 Elo / cost per task | Fable 5.1 Elo / cost per task |
|---|---|---|
| Low | 1224 / $0.21 | 1450 / $1.41 |
| Medium | 1576 / $0.86 | 1536 / $2.17 |
| High | 1692 / $1.54 | 1617 / $3.43 |
| Xhigh | 1820 / $4.21 | 1721 / $7.09 |
| Max | 1846 / $8.92 | 1735 / $9.59 |
Source: Anthropic’s knowledge-work chart.
Fable’s low-effort point is 226 Elo above Opus low, which is a reason not to assume the lowest settings are interchangeable. Opus medium, however, reaches 1576 at $0.86, above Fable low’s 1450 at $1.41. That supports comparing across settings as well as matching names.
At the other end, Opus xhigh is 85 Elo above Fable max, at $4.21 ÷ $9.59 = 43.9% of its displayed cost. Opus max extends its own rating by just 1846 − 1820 = 26 Elo, while cost rises by ($8.92 − $4.21) ÷ $4.21 = 111.9%. Elo differences cannot be converted into percentages of jobs saved, so that calculation does not tell you whether the upgrade pays on your documents.
Anthropic also reports a useful long-job example: both models translated HAProxy from C to Rust and passed nearly all of its regression tests. Opus finished in 9.5 hours versus Fable’s 12 and cost 51% less. The elapsed-time reduction is (12 − 9.5) ÷ 12 = 20.8%. This was Anthropic’s internal test, with no complete dollar bill in the launch account, so it remains a case study rather than a universal speed or cost estimate. Anthropic’s coding example.
Work through the bill before switching
The examples below use identical hypothetical token counts on Opus 5.5 and Fable 5.2. They isolate the rates and include reasoning in billed output. They exclude tools, taxes, regional premiums, fast mode and discounts. Cache-write tokens replace ordinary-input billing for those tokens; they are not counted in both categories. Real jobs will not use identical token counts: in Cursor’s table the two models use different amounts of work at the same effort label.
1. An ordinary uncached request
Assume 100,000 input tokens and 10,000 output tokens:
- Opus: 0.100 × $4 + 0.010 × $20 = $0.40 + $0.20 = $0.60.
- Fable 5.2: 0.100 × $8 + 0.010 × $40 = $0.80 + $0.40 = $1.20.
- Saving: ($1.20 − $0.60) ÷ $1.20 = 50%.
That is the headline rate advantage. Against Fable 5.1 the same request cost $1.50, and the saving was 60%.
2. A session with cache writes and repeated reads
Across many requests, assume 200,000 ordinary input tokens, 100,000 five-minute cache-write tokens, 1.8 million cache-read tokens and 150,000 output tokens. The input categories do not overlap:
- Opus: 0.2 × $4 + 0.1 × $5 + 1.8 × $0.20 + 0.15 × $20 = $0.80 + $0.50 + $0.36 + $3 = $4.66.
- Fable 5.2: 0.2 × $8 + 0.1 × $10 + 1.8 × $0.20 + 0.15 × $40 = $1.60 + $1 + $0.36 + $6 = $8.96.
- Saving: ($8.96 − $4.66) ÷ $8.96 = 48.0%.
If those writes use one-hour retention, replace Opus’s $0.50 write charge with 0.1 × $8 = $0.80, for $4.96 total. Replace Fable’s $1 with 0.1 × $16 = $1.60, for $9.56 total. The revised saving is ($9.56 − $4.96) ÷ $9.56 = 48.1%. Opus rate card, Fable 5.2 rate card, prompt-caching billing.
3. A cache-heavy follow-up
Consider a later request against an already warm cache: 100,000 cache-read tokens, 1,000 new input tokens and 1,000 output tokens.
- Opus: 0.100 × $0.20 + 0.001 × $4 + 0.001 × $20 = $0.020 + $0.004 + $0.020 = $0.044.
- Fable 5.2: 0.100 × $0.20 + 0.001 × $8 + 0.001 × $40 = $0.020 + $0.008 + $0.040 = $0.068.
- Saving: ($0.068 − $0.044) ÷ $0.068 = 35.3%.
This follow-up excludes the earlier cache fill. As cache reads take up more of an otherwise equal bill, the rate saving approaches zero, because both models charge $0.20 per million cache-read tokens. Opus remains cheaper at the listed rates, but “half the price” becomes a poor estimate of the saving on a cache-heavy session.
When Fable can still earn its premium
The published numbers give Fable 5.2 a stronger case than Fable 5.1 had. It holds the top score on Cursor’s table and leads most of Anthropic’s own rows, and the premium is smaller. Use it where you have evidence that it produces accepted work Opus misses, reduces expensive rework, or finishes long open-ended jobs that Opus does not. A claim that a task is “hard” is not enough on its own.
Using the session example above, Fable 5.2 costs $8.96 − $4.66 = $4.30 more per equal-token job. If a failed or deficient job costs $50 to repair, Fable would need to prevent more than $4.30 ÷ $50 = 8.6 percentage points of those repair events to repay the premium, assuming everything else is equal. That is a hypothetical break-even calculation; it is not an inferred failure-rate difference from a benchmark. With Fable 5.1 the same sum came to 13.08 points.
A cheaper-first policy can also be reasonable when failures are detectable. For 100 jobs at the stipulated session costs, all-Fable costs 100 × $8.96 = $896. Running all 100 on Opus and giving 20 a full additional Fable attempt costs 100 × $4.66 + 20 × $8.96 = $466 + $179.20 = $645.20. The difference is $250.80 ÷ $896 = 28.0% less. The 20% escalation rate is an assumption, and equal final quality has not been established. Review and handoff costs are extra.
Our cost-per-successful-job guide explains why that final qualification matters.
Switching models has a conversation-history cost
The model ID is only part of the switch. Anthropic’s migration guide says Fable 5.2 can read Opus 5.5’s thinking blocks on the Claude API, so a conversation that moves up from Opus to Fable keeps its earlier reasoning. The reverse does not hold. When a conversation moves from Fable 5.2 to Opus 5.5, the API drops Fable’s thinking blocks; the request still succeeds, and Anthropic says to expect higher cost and latency on the first turn after the switch. Start a fresh task with the relevant files, decisions and requirements when that is practical. Fable 5.2 migration guide.
Moving traffic from Opus 5.5 up to Fable 5.2 has its own checks. Fable 5.2 requires 30-day data retention and is not available under a zero-data-retention arrangement unless Anthropic authorizes it. Its rate limit is separate from Opus’s, with lower token limits at each tier. Its default effort is high where Opus’s is medium, so a request that omits effort runs one level higher. It also enforces the thinking-block prefix check for every account, which can break code that edits earlier messages between requests. Both models already require adaptive thinking and reject forced tool selection. Migrating from Opus 5.5.
If you are on Fable 5.1 today, Anthropic describes the move to 5.2 as mostly drop-in, with four changes that can break code: the prefix check, thinking blocks that Fable 5.1 cannot read back, a new computer-use toolset and stricter advisor pairing. Migrating from Fable 5.1.
Which should you choose?
| Situation | Recommendation |
|---|---|
| New coding workflow | Start with Opus medium; compare Opus high for harder repository tasks. |
| Coding budget under about $5 a task | Opus high. On Cursor’s table it outscores Fable 5.2 low and medium at a similar or lower cost. |
| Hardest coding work, where the top score matters | Test Fable 5.2 at high first, as Anthropic’s effort guide suggests; step up to xhigh or max only where you have measured a gain. |
| Documents, research and professional analysis | Start with Opus medium/high. Fable 5.2 leads most of Anthropic’s own knowledge-work and search rows, so test it on your own documents before paying for it. |
| Existing Fable 5.1 workflow | Plan the move to Fable 5.2: rates are 20% lower and Anthropic now lists 5.1 as legacy. Check the four breaking changes first. |
| Large, repeated cached context | Calculate the actual token mix; cache reads cost the same on both models, so Opus saves nothing on those tokens. |
| Conversation already well under way on Fable | Moving it to Opus drops Fable’s thinking blocks. Finish it on Fable or restart the task on Opus. |
| Zero-data-retention organization | Opus 5.5. Fable 5.2 needs 30-day retention unless Anthropic authorizes otherwise. |
I would still make Opus 5.5 medium/high the starting shortlist for new work, and I would now give Fable 5.2 a real trial on the hardest jobs. The published evidence supports that ordering. It does not establish how the two compare on your work, and Kingy has not yet run Fable 5.2 itself.
For the wider lineup, see our Fable 5.2 guide, Opus 5.5 launch review, Fable 5.1 review, Astra vs Opus effort comparison and Fable 5.2 vs GPT-6 Astra.
Sources and limits
Prices and specifications come from Anthropic’s model reference pages and pricing page, read November 9, 2026. The release date comes from Anthropic’s release notes. CursorBench values come from Cursor’s live table on the same day; Cursor can revise that table, as it did for Sonnet 5.5 pricing in October. The launch-table scores are the ones Anthropic published in its Fable 5.2 announcement. The Artificial Analysis index figures come from its model pages, read November 9, 2026; it can revise them.
The September section keeps its original sources: Terminal-Bench and GDPval values from Anthropic’s Opus 5.5 launch charts, read September 22, 2026. Those charts cover Fable 5.1 only. Anthropic’s benchmark notes describe production safeguards and fallback models on some evaluations, so the configuration is part of the result. The hypothetical bills above are labeled as such. No benchmark was rerun for this article, and no missing effort result was invented.
Update history
ShowHide
- November 9, 2026 (later): added Artificial Analysis’s independent index results for all three models.
- November 9, 2026: rebuilt around Claude Fable 5.2, released that day. New prices, CursorBench rows, Anthropic’s launch table, worked bills and switching notes. The Opus 5.5 vs Fable 5.1 charts moved to their own section.
- September 23, 2026: first published as an Opus 5.5 vs Fable 5.1 comparison.
