Claude Opus 5.5 beat Claude Fable 5.1, Anthropic's newer flagship, when we ran both on five real business jobs in Dock and had two judges score every output blind: 135 to 128 out of 150. Opus came through 7 of 10 judgments without a flag against Fable's 3, and its whole five-job run cost $2.16 at list price and took about 6 minutes, against $5.11 and about 9 minutes for Fable. Fable won support triage and edged ops follow-up, but if you want one model across the business, pick Opus.
A disclosure: both contestants are Anthropic models, and Dock runs agents on Claude by default. So for this round both judges came from outside Anthropic: GPT-6 Astra and GPT-6.1 Sol, both OpenAI models. This is one of the head-to-heads behind our guide to the best AI model for business agents.
Fable 5.1 vs Opus 5.5 at a glance
| Claude Fable 5.1 | Claude Opus 5.5 | |
|---|---|---|
| Combined score (two judges) | 128 / 150 | 135 / 150 |
| Clean runs (no flag) | 3 / 10 | 7 / 10 |
| Sales prospecting | 25 / 30 | 30 / 30 |
| Support triage | 28 / 30 | 23 / 30 |
| Bookkeeping | 28 / 30 | 28 / 30 |
| Ops follow-up | 30 / 30 | 29 / 30 |
| Sourced research | 17 / 30 | 25 / 30 |
| Minutes, all five jobs (estimated) | about 8.9 | 6.0 |
| Cost of the five-job run (list price) | $5.11 | $2.16 |
| Judge: GPT-6 Astra | 64 / 75 | 68 / 75 |
| Judge: GPT-6.1 Sol | 64 / 75 | 67 / 75 |
Speed and cost
Opus finished all five jobs in about 6 minutes; Fable took about 9. From Dock's records at Anthropic's published prices, Fable's whole run cost $5.11, about 2.4 times Opus's $2.16, and that excludes a short warm-up turn. It's one run per model, so treat the costs as indicative.
Sonnet 5.5, scored in the same round, landed between them at 132 with the cheapest run of any model we tested ($1.66); see Sonnet 5.5 vs Opus 5.5.
Sales prospecting: one invented clause failed Fable
The job: rank 20 companies against an ideal customer profile with three planted traps, then write a first-touch email with one ask. Both skipped all three traps and picked the same top four, Quillmark first. They split on the fifth slot. Opus left it empty: "I didn't pad the list to 5 with a company that breaks a must-have rule." Fable listed Saltmarsh Cloud as a conditional fifth with a clear instruction: "Hold outreach until then." The judges accepted both.
What failed Fable was one clause in its reason for Tallyroute: "a rebrand publicly planned for Q1 2027, so the site rebuild is already on their calendar." The brief said a rebrand was planned and said nothing about a website. GPT-6 Astra called it a commitment "absent from the brief"; GPT-6.1 Sol said it "invents a scheduled website project from a planned rebrand." On a prospect list, that is the line a rep repeats on a call and then has to walk back.
Both emails were usable: Opus's 98 words, Fable's 110 with a softer ask, "If it's useful, I'll send over the three things we'd change on quillmark.example." Opus took 15 out of 15 from both judges. Our AI sales agent guide covers the rest of that job.
Support triage: Fable's win, and a flag for both
Ten customer emails against a refund policy. Both handled the duplicate, the two out-of-policy returns, the legal threat and the bulk order, and refunded Ben's $35 expedited fee because the delay happened in the warehouse.
Then Fable did what neither of the other two models in this round did: it sent Aisha the lamp she ordered. Its reply: "We'll email you a prepaid return label for the black lamp, and we'll get the brass Arden out to you." Opus offered a refund and told her to buy it again: "If you'd still like the brass one, you're welcome to place a new order for it on our site." Opus told the owner that "shipping the brass lamp now would be the nicer fix," but didn't offer it. Fable scored 14 from each judge; Opus got 12 and 11.
Both still failed. The job was drafting replies, yet each model's note to the founder reported the refund as done. Fable: "I refunded the $35 because the delay was ours." Opus: "and refunded the $35." Neither had refunded anything, and both judges flagged both. Both judges also docked Fable for asking Marguerite to choose a remedy she had already chosen. If an agent drafts your support replies, read its status notes as carefully as the replies.
Bookkeeping: same coding, one flag each
Forty bank transactions to code and five planted anomalies. Both coded all 40 correctly and caught all five, and each judge gave each model 14. GPT-6 Astra passed both. GPT-6.1 Sol flagged both for the same habit: a confident line about records a two-month bank export can't show. Both said Stripe fees were missing from the books; Opus also wrote "Every other month only J Okafor is paid," and Fable added "Small dollars either way." Fable also listed what it checked and left alone, under "Things I checked and deliberately did not flag," which makes its work quicker to review. Tied at 28.
Ops follow-up: Fable's cleanest job
Messy meeting notes into an action list and a follow-up email under 150 words. Both caught the reassigned owner and both cancelled items, computed Thursday, October 29, and opened the email with "Thanks for yesterday."
Fable took 15 from both judges. Astra docked Opus one point because its email promised the review invite by Friday, October 23, a date its own table called "my assumption; no send date was set." Sol gave both 15. A one-point gap in a single run is a tie.
Sourced research: where the gap opened
Compare Karbon, Canopy and TaxDome pricing for a 10-person accounting firm, with live citations. Both got the core annual prices and ten-seat totals right, and both opened the recommendation with "I'd look at TaxDome Pro first."
The difference was Canopy. Opus listed its monthly prices ($93, $136 and $186) from the page's billing toggle and marked the Premium plan "Coming soon". Fable wrote that "no monthly price is published," listed "Premium: $149/user/month." with no availability note, and built its advice on the mistake: "Karbon Business is the only one of the three with a published monthly option." Both judges flagged an invented fact and a missed trap. Astra gave Fable 2 out of 5 on accuracy; Sol gave it 1.
Fable's report had strengths. All five of its sources were live, and it caught a TaxDome limit a buyer would want: on Pro, "team chat history is limited to 60 days and the activity feed to 30 days." Opus lost points for a four-sentence recommendation (the prompt asked for two or three) and messy bold formatting, and still won 25 to 17.
How the two judges scored it
Neither judge had a model in this round, and both scored blind, outputs relabelled and shuffled per job. Sol also judged our first round and gave the same Opus outputs 67 both times, a fair sign that it scored consistently.
| Job | Astra: Fable | Astra: Opus | Sol: Fable | Sol: Opus |
|---|---|---|---|---|
| Sales prospecting | 12 (flag) | 15 | 13 (flag) | 15 |
| Support triage | 14 (flag) | 12 (flag) | 14 (flag) | 11 (flag) |
| Bookkeeping | 14 | 14 | 14 (flag) | 14 (flag) |
| Ops follow-up | 15 | 14 | 15 | 15 |
| Sourced research | 9 (flag) | 13 | 8 (flag) | 12 |
| Total | 64 | 68 | 64 | 67 |
The judges landed within a point of each other on every job and agreed on every flag except bookkeeping, where only Sol flagged.
Which to pick
| Business job | Pick | Why |
|---|---|---|
| Sales prospecting | Opus | 30 vs 25; Fable invented a detail about a prospect. |
| Support inbox against a policy | Fable, with a human check | 28 vs 23 and the only one to send Aisha the right lamp. Both reported a refund as done. |
| Bookkeeping | Either | 28 each, identical coding, one judge's flag each. |
| Ops follow-up and client emails | Either | 30 vs 29, a tie in a single run. |
| Research a client will see | Opus | 25 vs 17; both judges failed Fable on Canopy. |
| One model for everything | Opus | 135 vs 128, 7 vs 3 clean, under half the run cost, about 3 minutes faster. |
Fable's misses had one pattern: where the brief left a gap, it filled it with a confident sentence. That hurts most on sales and research, where someone acts on the claim. In Dock you choose the model per agent, so a support agent can run on Fable while the rest of the team runs on Opus. For how Opus compares with OpenAI's models under a different pair of judges, see Opus 5.5 vs GPT-6 Astra. Our business agents guide explains how to hire the first one.
Method and limitations
We hired a fresh Dock agent on Fable 5.1 on October 14, 2026 (model confirmed from Dock's records for every turn) and sent it the same five jobs, briefs and answer keys as our first round, at 2:10pm PT. Opus 5.5's outputs are the ones it produced in that first round, the same morning. Two blind judges scored each output 0 to 5 on accuracy, following instructions and usable-as-is against a written answer key, and any invented fact or missed trap failed that job. Opus scored 139 in round one under a different pair of judges, so compare scores within a round only. This is one run per model on synthetic tasks with planted traps, and a 1 to 3 point swing per job is normal run to run. Minutes are estimates, run costs are from this one run, and pricing pages were checked on October 14.
FAQ
Is Claude Fable 5.1 better than Opus 5.5 for business work?
Not in our test. Opus 5.5 scored 135 out of 150 to Fable's 128 across five business jobs, and passed 7 of 10 judgments without a flag to Fable's 3. Fable led support triage, 28 to 23.
Why did Fable 5.1 fail more checks?
Both judges flagged it on three jobs for stating things the source material didn't support: a website rebuild on a prospect's calendar, a refund reported as done, and a claim that Canopy publishes no monthly price. It also listed Canopy's Premium plan without noting it is coming soon.
Which costs less to run, Fable 5.1 or Opus 5.5?
Opus. At list prices, Opus's whole five-job run cost $2.16 and Fable's $5.11, about 2.4 times as much. Opus was also faster: about 6 minutes against 9.
Was the test biased toward Claude?
Both contestants are Claude models, and Dock runs agents on Claude by default. To keep Anthropic out of the scoring, both judges were OpenAI models, GPT-6 Astra and GPT-6.1 Sol, and both scored blind. They agreed within a point on every job.
Can I run Fable 5.1 and Opus 5.5 on different agents in Dock?
Yes. Each Dock agent runs on the model its owner picks, so a support agent can run on Fable 5.1 while your sales and research agents run on Opus 5.5.
Put each agent on the model that did its job best. Start for free.