AI for performance marketing
How Anthropic's Claude Opus 5.5 performs on 16 performance-marketing tasks, from ad concepts to campaign readouts, and what it costs per output.
On Benmark's 16 performance-marketing tasks, Claude Opus 5.5 has the outright top score for concepting from a brand brief (8.11, against Gemini 4 Carbon's 8.00) and the second-highest ideation average (8.23, after Carbon's 8.30). It is level with Gemini 4 Argon (6 wins, 6 losses and 4 ties) at a median of about 13 times Argon's cost per output. It is a little behind Claude Fable 5.2 (4 wins, 6 losses and 6 ties), which costs a median 1.3 times more. Its weakest stage is briefing: hooks (6.75, joint 18th of 21) and narrative (7.75, against 9.00 at the top), both from one paired observation.
Average across all 16 tasks: 8.70, fourth of the 12 models tested on every task.
Opus 5.5 is still my pick for turning a bare brand brief into concepts. Elsewhere it's close to Fable 5.2 or behind it, at a similar estimated cost, and its hooks are weak. Overall it is a strong model, however it doesn't win individually in many areas with other models either outperforming on tasks, or undercutting costs at pretty much the same quality.
Concepting from a brand brief: the top score. Opus 5.5 has the outright top score for concepting from a brand brief (8.11, three paired observations) and the second-highest ideation average (8.23, after Gemini 4 Carbon's 8.30).
Briefing: use another model. Opus 5.5 ranks fifteenth of 18 on the briefing average (8.00). Its hooks (6.75) are joint 18th of 21 and its narrative (7.75) trails Fable 5.2's 9.00, both from one paired observation.
Launch and analysis: strong, but Argon is cheaper. Opus 5.5 has the third-highest launch average (9.36, after Claude Fable 5.2's 9.52 and Gemini 4 Carbon's 9.37) and ranks fifth on analysis (9.59). Gemini 4 Argon beats it on performance readouts, ad analysis, content review and creative tagging, at 7 to 20 times lower cost per output.
Simple prompt for readouts and proposals. Opus 5.5's performance readouts drop from 8.86 to 8.07 with the scaffold (7 paired observations), and its insight proposals from 7.67 to 7.22 (3). The scaffold helps insight codification: 5.87 to 7.56 (2).
Against Gemini 4 Argon, rarely. They're level overall, and on the 7 tasks with comparable costs for both, Opus 5.5's estimated cost per output is a median 13 times Argon's recorded spend. The case for Opus 5.5 is concepting from a brand brief, where it leads Argon by 0.89, but it has no comparable cost on 3 of the 4 ideation tasks. Against Claude Fable 5.2 it's slightly cheaper: Fable 5.2 costs a median 1.3 times more, on 6 tasks estimated the same way.
| Compared with | Claude Opus 5.5 ahead | Behind | Level | Claude Opus 5.5’s cost per output |
|---|---|---|---|---|
| Gemini 4 Carbon | 3 | 8 | 5 | 5.4× higher (median of 7 tasks) |
| Claude Fable 5.2 | 4 | 6 | 6 | 1.3× lower (median of 6 tasks) |
| Gemini 4 Argon | 6 | 6 | 4 | 13× higher (median of 7 tasks) |
| GPT-6 Astra | 10 | 2 | 4 | 3.6× higher (median of 7 tasks) |
| Claude Opus 5 | 8 | 5 | 3 | 4.1× higher (median of 7 tasks) |
Each row compares published scores task by task, with each model on its better prompt setup; many margins are small. Costs for Claude Opus 5.5 and Claude Fable 5.2 are API-equivalent estimates. Costs for Gemini 4 Carbon, Gemini 4 Argon, GPT-6 Astra and Claude Opus 5 are recorded spend. Gemini 4 Carbon's and Gemini 4 Argon's costs were recorded at their introductory prices, which end on 30 November 2026. Mixed-basis costs are excluded.
Carbon wins 8 tasks, Opus 5.5 wins 3, and 5 are level. Opus 5.5's wins are narrow: concepting from a brand brief (8.11 against 8.00), iterating on a top performer (9.43 against 9.20) and ad analysis (9.90 against 9.83). Its estimated cost per output is a median 5.4 times Carbon's recorded spend.
Fable 5.2 wins 6, Opus 5.5 wins 4, and 6 are level. Opus 5.5 leads on concepting from a brand brief (8.11 against 7.33). Fable 5.2 is far ahead on hooks (8.50 against 6.75) and narrative (9.00 against 7.75), both from one paired observation, and leads on asset naming (9.12 against 8.65). Fable 5.2 costs a median 1.3 times more per output.
Each wins 6 tasks, and 4 are level. Opus 5.5 leads on concepting from a brand brief (8.11 against 7.22) and asset naming (8.65 against 8.21). Argon leads on hooks (8.25 against 6.75, one paired observation) and adapting a reference ad (8.33 against 7.72). Opus 5.5 costs a median 13 times more per output.
Opus 5.5 wins 10 and loses 2, hooks (6.75 against 8.00) and creative tagging (9.43 against 9.47), with 4 level. Its biggest leads are content review (7.79 against 5.12, 20 paired observations) and concepting from a brand brief (8.11 against 6.11). Opus 5.5 costs a median 3.6 times more per output.
On most tasks. It wins 8, loses 5 and ties 3. The biggest gains are concepting from a brand brief (5.78 to 8.11) and content review (6.24 to 7.79). The biggest drops are hooks (8.50 to 6.75) and narrative (8.50 to 7.75), both from one paired observation. Its estimated cost per output is a median 4.1 times Opus 5's recorded spend.
| Stage | Task | Score | Prompt | Rank | Sample | Gap to top | Cost per output | Status |
|---|---|---|---|---|---|---|---|---|
| Ideation | ||||||||
| Ideation | Reference ad to new concept | 7.72 | Simple | 7 of 18 | 6 paired | −0.61 | Mixed basis | Provisional |
| Ideation | Brand context to concepts | 8.11 | Simple | 1 of 18 | 3 paired | Top | Mixed basis | Provisional |
| Ideation | Insights and ads to a new proposal | 7.67 | Simple | 3 of 18 | 3 paired | −0.33 | Mixed basis | Provisional |
| Ideation | Iterate on a top performer | 9.43 | Scaffolded | =2 of 15 | 10 paired | −0.04 | $0.573 | Provisional |
| Planning | ||||||||
| Planning | Idea scoring and priority plan | 10.00 | Scaffolded | =1 of 19 | 1 paired | Top (joint) | $1.626 | Provisional |
| Briefing | ||||||||
| Briefing | Hook development | 6.75 | Simple | =18 of 21 | 1 paired | −2.00 | Not recorded | Published |
| Briefing | Narrative development | 7.75 | Simple | =11 of 21 | 1 paired | −1.25 | Not recorded | Published |
| Briefing | Compliant brief generation | 9.50 | Simple | =1 of 18 | 1 paired | Top (joint) | Not recorded | Published |
| Production | ||||||||
| Production | Content review | 7.79 | Simple | 4 of 18 | 20 paired | −0.67 | $0.239 | Provisional |
| Launch | ||||||||
| Launch | Creative tagging | 9.43 | Combined | =8 of 19 | 20 outputs | −0.30 | $0.339 | Provisional |
| Launch | Asset naming and versioning | 8.65 | Scaffolded | =2 of 18 | 7 outputs | −0.47 | Mixed basis | Provisional |
| Launch | Deployment QA checklist | 10.00 | Combined | =1 of 17 | 2 outputs | Top (joint) | $0.945 | Provisional |
| Analysis | ||||||||
| Analysis | Performance readout | 8.86 | Simple | 5 of 20 | 7 paired | −0.50 | $2.041 | Provisional |
| Analysis | Why an ad worked | 9.90 | Scaffolded | =3 of 15 | 10 paired | −0.03 | $0.622 | Provisional |
| Analysis | Creative diversity and coverage | 10.00 | Simple | =1 of 18 | 2 paired | Top (joint) | Mixed basis | Provisional |
| Insight | ||||||||
| Insight | Insight codification | 7.56 | Scaffolded | 6 of 19 | 2 paired | −0.90 | Mixed basis | Provisional |
| Stage | Tasks | Claude Opus 5.5 average | Rank | Stage leader |
|---|---|---|---|---|
| Ideation | 4 | 8.23 | 2 of 14 | Gemini 4 Carbon (8.30) |
| Planning | 1 | 10.00 | =1 of 19 | 12 models tied (10.00) |
| Briefing | 3 | 8.00 | 15 of 18 | Claude Fable 5.2 (9.00) |
| Production | 1 | 7.79 | 4 of 18 | Gemini 4 Carbon (8.46) |
| Launch | 3 | 9.36 | 3 of 16 | Claude Fable 5.2 (9.52) |
| Analysis | 3 | 9.59 | 5 of 14 | Claude Fable 5.1 (9.75) |
| Insight | 1 | 7.56 | 6 of 19 | Gemini 4 Carbon (8.46) |
Averages are unweighted means of the published task scores. Rank counts only models with a score on every task in the stage. Stages cover the tasks in Full results on this page.
Every model gets the same fixed inputs for an invented brand: reference ads, brand briefs, campaign data or asset records, depending on the task. Each input runs with a simple prompt and with a scaffolded prompt that adds a method and criteria, and a model's published score is its better setup. Creative work is blind-scored against a skilled-human standard; structured tasks are scored automatically against answer keys. Each task page sets out its inputs, prompts and scoring. Claude Opus 5.5 was tested in September 2026 and October 2026.
Benmark is an independent benchmark of AI models on performance-marketing work, run by Ben, who co-founded and sold a performance marketing agency. No model provider pays for or reviews these results.
Benmark (2026). Claude Opus 5.5 for marketing: benchmark results across 16 tasks. https://benmark.ai/models/claude-opus-5-5/
Useful, occasional updates
New benchmark results and lab experiments when there’s something worth sharing. No generic newsletter.
Follow the experiments →Ask Ben anything ↗