Compare / Qwen3.8-Max (0902) vs Claude Haiku 5.5
Qwen3.8-Max (0902) vs Claude Haiku 5.5
Which is cheaper, which is more capable, and which does better at each kind of work, from 62 results on 4 sources that measured both models.
Results as of 8 October 2026 (catalogue release 2026-10-08-4199ce0620fe).
Qwen3.8-Max (0902) or Claude Haiku 5.5? The short answer
They score about the same, so pick Claude Haiku 5.5: it costs 21 times less per product listing on CatalogBench.
- Spring Prompt overall: Qwen3.8-Max (0902) 48, Claude Haiku 5.5 44 (out of 100; how the score works).
- Claude Haiku 5.5 is cheaper to run: $0.001 against $0.026 per product listing on CatalogBench (21 times less).
- Qwen3.8-Max (0902) scores higher on the Artificial Analysis Intelligence Index: 45.4 against 43.4.
- Claude Haiku 5.5 writes faster: 235 tokens a second against 37.
- Claude Haiku 5.5 is newer: released 7 Oct 2026, against 3 Aug 2026 for Qwen3.8-Max (0902).
- On our own benchmarks, on the results both models have: Qwen3.8-Max (0902) does better for marketing planning.
- On third-party benchmarks, on the results both models have: Qwen3.8-Max (0902) does better for agents and tool use.
At a glance
| Qwen3.8-Max (0902) | Claude Haiku 5.5 | |
|---|---|---|
| Price per million tokens | $2.00 in · $6.00 out | $0.10 in · $0.50 out |
| Blended (3 in : 1 out) | $3.00 | $0.20 |
| Cheapest host, blended | – | – |
| Cost per product listing (CatalogBench) | $0.026 | $0.001 |
| Intelligence Index | 45.4 | 43.4 |
| Spring Prompt overall | 48 | 44 |
| Output speed, tokens a second | 37 | 235 |
| Context window | 1,000,000 tokens | 1,000,000 tokens |
| Released | 3 Aug 2026 | 7 Oct 2026 (newer) |
| Weights | Closed | Closed |
| Developer based in | China | the US |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among models available today, for each kind of work, from our own benchmarks and licensed sources. The better rank is in green.
| Use case | Qwen3.8-Max (0902) | Claude Haiku 5.5 |
|---|---|---|
| Product listingsTurning a sparse product feed and photos into listings that can go live | 17 of 20 | 11 of 20 |
| Decks from an analysisTurning a finished analysis into a deck you could present as it is | 11 of 19 | 10 of 19 |
| User surveysPlanning a user survey and reading its results without being misled | not measured | 3 of 8 |
| Marketing planningPlanning a year of ad spend without overspending | 14 of 19 | 17 of 19 |
| Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls | 7 of 164 | 20 of 164 |
| Professional workReal tasks from banking, consulting and law, business documents and freelance projects | 18 of 51 | not measured |
| Reasoning and knowledgeHard questions across science, maths and general knowledge | 17 of 198 | 19 of 198 |
| WritingWhat people prefer in blind comparisons, and judged writing quality | 16 of 188 | not measured |
| Sticking to the factsSummarising without inventing things, and factual answers | 38 of 152 | not measured |
| Speed under pressureGood decisions against a real clock (fast chess) | 23 of 23 | 15 of 22 |
From our research
Every shared result
Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.
| Measure | Qwen3.8-Max (0902) | Claude Haiku 5.5 |
|---|---|---|
| CatalogBench measured by Spring Prompt · 3 results | ||
| Cost per productUS dollars, lower is better | $0.0256 | $0.0012 |
| Reliably publish-ready% of products, higher is better | 28.6% | 26.8% |
| Reliably publish-ready · sales brief% of products, higher is better | 0.0% | 10.7% |
| DeckBench measured by Spring Prompt · 3 results | ||
| Cost per deckUS dollars, lower is better | $0.15 | $0.0091 |
| Deck ratingrating, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. | 1,100 | 1,130 |
| Accurate decks% of tasks, higher is better | 50.0% | 33.3% |
| ROASBench measured by Spring Prompt · 3 results | ||
| Overall scorescore out of 100, higher is better | 25.4 | 24.1 |
| Audience scorescore out of 100, higher is better | 39.7 | 48.4 |
| Business scorescore out of 100, higher is better | 15.6 | 13.5 |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better | 45.4 | 43.4max reasoning |
| Humanity's Last Exam% of questions, higher is better | 43.1% | 44.4%max reasoning |
| Terminal-Bench 4.0% of tasks, higher is better | 38.9% | 32.8%max reasoning |
Show all 62 results (50 more)
| Measure | Qwen3.8-Max (0902) | Claude Haiku 5.5 |
|---|---|---|
| CatalogBench measured by Spring Prompt · 27 results | ||
| Channel compliance% of products, higher is better | 100.0% | 100.0% |
| Channel rules broken% of products, lower is better | 0.6% | 0.6% |
| Channel rules broken · sales brief% of products, lower is better | 3.0% | 0.0% |
| Claims to check% of products, lower is better | 0.0% | 0.6% |
| Claims to check · sales brief% of products, lower is better | 1.3% | 1.2% |
| Conflicts caught% of conflicts, higher is better | 100.0% | 100.0% |
| Content quality% of checks, higher is better | 94.2% | 95.6% |
| Content quality · sales brief% of checks, higher is better | 93.6% | 96.3% |
| Decision accuracy% of decisions, higher is better | 93.3% | 95.6% |
| Failed outputs% of products, lower is better | 3.6% | 0.0% |
| Failed outputs · sales brief% of products, lower is better | 3.6% | 0.0% |
| Field accuracy% of missing fields, higher is better | 90.3% | 91.0% |
| Invented values% of filled values, lower is better | 2.7% | 0.8% |
| Missing UK information% of products, lower is better | 0.0% | 0.0% |
| Missing UK information · sales brief% of products, lower is better | 0.0% | 0.0% |
| Not findable% of products, lower is better | 3.6% | 11.9% |
| Not findable · sales brief% of products, lower is better | 1.8% | 10.7% |
| Publish-ready listings% of products, higher is better | 50.0% | 47.0% |
| Publish-ready listings · sales brief% of products, higher is better | 7.7% | 20.2% |
| Unsupported claims% of products, lower is better | 26.8% | 32.1% |
| Unsupported claimsclaims per product, lower is better | 0.69 | 1.20 |
| Unsupported claims · sales brief% of products, lower is better | 84.5% | 70.2% |
| Unsupported claims · sales briefclaims per product, lower is better | 4.52 | 2.46 |
| Wrong attributes% of products, lower is better | 21.4% | 19.6% |
| Wrong attributes · sales brief% of products, lower is better | 19.6% | 18.5% |
| Wrong category or variant% of products, lower is better | 1.8% | 0.0% |
| Wrong category or variant · sales brief% of products, lower is better | 2.4% | 0.0% |
| DeckBench measured by Spring Prompt · 13 results | ||
| Caveat dropped% of tasks, lower is better | 16.7% | 16.7% |
| Clean layout% of tasks, higher is better | 66.7% | 66.7% |
| Design quality% of the maximum, higher is better | 67.0% | 61.1% |
| Draft figure quoted% of tasks, lower is better | 33.3% | 33.3% |
| Findings missing% of tasks, lower is better | 0.0% | 16.7% |
| Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. | 61.5% | 50.0% |
| Layout defects% of tasks, lower is better | 33.3% | 33.3% |
| Misleading metric used% of tasks, lower is better | 0.0% | 16.7% |
| Presentable decks% of tasks, higher is better | 33.3% | 33.3% |
| Recommendation late or wrong% of tasks, lower is better | 0.0% | 33.3% |
| Slides needing work% of tasks, lower is better | 50.0% | 16.7% |
| Unsupported claims% of tasks, lower is better | 16.7% | 33.3% |
| Unsupported numbers% of tasks, lower is better | 0.0% | 16.7% |
| ROASBench measured by Spring Prompt · 6 results | ||
| Consistency scorescore out of 100, higher is better | 19.3 | 10.6 |
| Contribution profitsimulated US dollars, higher is better | $99,775 | $32,146 |
| Cost of a runUS dollars, lower is better | $0.73 | $0.0313 |
| Months over budgetmonths of 12, lower is better | 2 | 0 |
| Planning scorescore out of 100, higher is better | 54.8 | 54.8 |
| Return on ad spendprofit per $1 spent, higher is better | $0.09 | $0.03 |
| Artificial Analysis reported by Artificial Analysis · 4 results | ||
| Long-context reasoning (AA-LCR)% of questions, higher is better | 80.3% | 82.7%max reasoning |
| Output speedtokens per second, higher is better | 37.0 | 235max reasoning |
| SciCode% of problems, higher is better | 52.1% | 55.0%max reasoning |
| Time to first answer tokenseconds, lower is better | 55.8 | 5.08low reasoning |
Quick answers
Is Qwen3.8-Max (0902) better than Claude Haiku 5.5?
They score about the same, so pick Claude Haiku 5.5: it costs 21 times less per product listing on CatalogBench.
Which is cheaper, Qwen3.8-Max (0902) or Claude Haiku 5.5?
Claude Haiku 5.5 is cheaper to run: $0.001 against $0.026 per product listing on CatalogBench (21 times less).
Which is faster, Qwen3.8-Max (0902) or Claude Haiku 5.5?
Claude Haiku 5.5 writes about 235 tokens a second on its usual API, against 37 for Qwen3.8-Max (0902) (Artificial Analysis).
Run Qwen3.8-Max (0902) and Claude Haiku 5.5 on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.