Compare / Qwen3.5-35B-A3B vs Claude Haiku 4.5

Qwen3.5-35B-A3B vs Claude Haiku 4.5

Which is cheaper, which is more capable, and which does better at each kind of work, from 22 results on 4 sources that measured both models.

AlibabaQwen3.5-35B-A3B Released 24 Feb 2026 · $0.16 in · $1.30 out
AnthropicClaude Haiku 4.5 Released 15 Oct 2025 · $1.00 in · $5.00 out
Qwen3.5-35B-A3B wins 6Claude Haiku 4.5 wins 16

Of the 22 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.

Results as of 9 October 2026 (catalogue release 2026-10-09-dd620e778ac5). See also Alibaba vs Anthropic, every model from both labs.

Qwen3.5-35B-A3B or Claude Haiku 4.5? The short answer

Pick Qwen3.5-35B-A3B: it scores higher (Intelligence Index 19.3 against 16.9) and costs less.

  • Qwen3.5-35B-A3B is cheaper: $0.45 against $2.00 per million tokens, 4.5 times less.
  • Qwen3.5-35B-A3B scores higher on the Artificial Analysis Intelligence Index: 19.3 against 16.9.
  • Qwen3.5-35B-A3B is newer: released 24 Feb 2026, against 15 Oct 2025 for Claude Haiku 4.5.
  • On third-party benchmarks, on the results both models have: Qwen3.5-35B-A3B does better for reasoning and knowledge; Claude Haiku 4.5 for agents and tool use, writing and sticking to the facts.

Head to head

19.3
Intelligence Index
16.9
$0.45
Price per million tokens, blendedlower is better
$2.00
262,144
Context window, tokens
200,000

Every fact side by side

Qwen3.5-35B-A3BClaude Haiku 4.5
Price per million tokens$0.16 in · $1.30 out$1.00 in · $5.00 out
Blended (3 in : 1 out)$0.45$2.00
Cheapest host, blended$0.25 (Darkbloom, FP4)$2.00 (Amazon Bedrock)
Cached input, per million–$0.10 cached input
Intelligence Index19.316.9
Spring Prompt overall–7
Output speed, tokens a second–97
Context window262,144 tokens200,000 tokens
Released24 Feb 2026 (newer)15 Oct 2025
WeightsOpenClosed
Developer based inChinathe 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 the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.

Product listingsTurning a sparse product feed and photos into listings that can go live not measured 19 of 20
Decks from an analysisTurning a finished analysis into a deck you could present as it is not measured 15 of 19
User surveysPlanning a user survey and reading its results without being misled not measured 8 of 8
Marketing planningPlanning a year of ad spend without overspending not measured 18 of 19
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 116 of 164 91 of 164
Reasoning and knowledgeHard questions across science, maths and general knowledge 92 of 198 127 of 198
WritingWhat people prefer in blind comparisons, and judged writing quality 108 of 188 87 of 188
Sticking to the factsSummarising without inventing things, and factual answers 112 of 152 117 of 152
Speed under pressureGood decisions against a real clock (fast chess) not measured 17 of 23

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. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. 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.

MeasureQwen3.5-35B-A3BClaude Haiku 4.5
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results
Overall · TextArena rating, higher is better 147 · 121 of 387 measured 1,394 1,414
Overall · Text factualityArena rating, higher is better 129 · 118 of 163 measured 1,407 1,418
Business, management and finance · TextArena rating, higher is better 139 · 109 of 380 measured 1,403 1,420
Artificial Analysis reported by Artificial Analysis · 3 results
Artificial Analysis Intelligence Indexindex score, higher is better 108 · 116 of 243 measured 19.3 16.9thinking reasoning
GPQA Diamond% of questions, higher is better 84 · 163 of 231 measured 84.5% 67.2%thinking reasoning
Humanity's Last Exam% of questions, higher is better 106 · 149 of 242 measured 21.0% 10.4%thinking reasoning
UGI Leaderboard reported by UGI Leaderboard · 3 results
Writing score · UGIscore out of 100, higher is better 97 · 72 of 230 measured 42.6think-prefill reasoning 52.2no reasoning
Requested-length error · UGI% off the requested word count, lower is better 83 · 17 of 230 measured 17.0%no reasoning 6.0%no reasoning
Style adherence · UGIscore from 0 to 1, higher is better 151 · 22 of 230 measured 0.33no reasoning 0.39thinking reasoning
Vectara Hallucination Leaderboard reported by Vectara Hallucination Leaderboard · 2 results
Hallucination rate · Vectara% of summaries, lower is better 62 · 58 of 104 measured 10.5% 9.8%
Answer rate · Vectara% of documents, higher is better 31 · 50 of 104 measured 99.8% 99.5%
Show all 22 results (11 more)
MeasureQwen3.5-35B-A3BClaude Haiku 4.5
Arena (formerly LMArena) reported by Arena (formerly LMArena) · 4 results
Creative writing · TextArena rating, higher is better 163 · 112 of 385 measured 1,342 1,388
Expert prompts · TextArena rating, higher is better 134 · 97 of 338 measured 1,419 1,451
Instruction following · TextArena rating, higher is better 140 · 101 of 387 measured 1,386 1,415
Writing, literature and language · TextArena rating, higher is better 155 · 106 of 386 measured 1,363 1,400
Artificial Analysis reported by Artificial Analysis · 7 results
Terminal-Bench 2.1% of tasks, higher is better 72 · 70 of 130 measured 40.8%no reasoning 44.2%thinking reasoning
τ-bench banking% of tasks, higher is better 107 · 82 of 121 measured 4.9%no reasoning 9.3%thinking reasoning
Artificial Analysis Coding Indexindex score, higher is better 79 · 68 of 131 measured 37.0no reasoning 43.9thinking reasoning
IFBench% of instructions, higher is better 39 · 92 of 195 measured 72.5% 54.3%thinking reasoning
Long-context reasoning (AA-LCR)% of questions, higher is better 99 · 82 of 233 measured 72.0% 74.3%thinking reasoning
Terminal-Bench Hard% of tasks, higher is better 74 · 72 of 190 measured 26.5% 27.3%no reasoning
τ²-bench telecom% of tasks, higher is better 45 · 98 of 193 measured 89.2% 54.7%thinking reasoning

Quick answers

Is Qwen3.5-35B-A3B better than Claude Haiku 4.5?

Pick Qwen3.5-35B-A3B: it scores higher (Intelligence Index 19.3 against 16.9) and costs less.

Which is cheaper, Qwen3.5-35B-A3B or Claude Haiku 4.5?

Qwen3.5-35B-A3B is cheaper: $0.45 against $2.00 per million tokens, 4.5 times less.

Which is better for agents and tool use, Qwen3.5-35B-A3B or Claude Haiku 4.5?

Claude Haiku 4.5 ranks 91 of 164 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 116 for Qwen3.5-35B-A3B.

Which is better for reasoning and knowledge, Qwen3.5-35B-A3B or Claude Haiku 4.5?

Qwen3.5-35B-A3B ranks 92 of 198 models on sale for hard questions across science, maths and general knowledge, against 127 for Claude Haiku 4.5.

Which is better for writing, Qwen3.5-35B-A3B or Claude Haiku 4.5?

Claude Haiku 4.5 ranks 87 of 188 models on sale for what people prefer in blind comparisons, and judged writing quality, against 108 for Qwen3.5-35B-A3B.

Which is better for sticking to the facts, Qwen3.5-35B-A3B or Claude Haiku 4.5?

Qwen3.5-35B-A3B ranks 112 of 152 models on sale for summarising without inventing things, and factual answers, against 117 for Claude Haiku 4.5.

Which has the bigger context window, Qwen3.5-35B-A3B or Claude Haiku 4.5?

Qwen3.5-35B-A3B: 262,144 tokens, against 200,000 for Claude Haiku 4.5.

Run Qwen3.5-35B-A3B and Claude Haiku 4.5 on your own prompt

Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.

Try the playgroundTalk to us