Compare / Qwen3-8B vs Claude Sonnet 4.5
Qwen3-8B vs Claude Sonnet 4.5
Which is cheaper, which is more capable, and which does better at each kind of work, from 25 results on 4 sources that measured both models.
Of the 25 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 14 October 2026 (catalogue release 2026-10-14-7d7cf32c3f6d). See also Alibaba vs Anthropic, every model from both labs.
Qwen3-8B or Claude Sonnet 4.5? The short answer
Pick Claude Sonnet 4.5 if quality matters most: it scores higher (Intelligence Index 20.7 against 7.3). Pick Qwen3-8B if cost matters more: it costs 30 times less per token.
- Qwen3-8B is cheaper: $0.20 against $6.00 per million tokens, 30 times less.
- Claude Sonnet 4.5 scores higher on the Artificial Analysis Intelligence Index: 20.7 against 7.3.
- Claude Sonnet 4.5 is newer: released 29 Sep 2025, against 28 Apr 2025 for Qwen3-8B.
- On third-party benchmarks, on the results both models have: Qwen3-8B does better for sticking to the facts; Claude Sonnet 4.5 for agents and tool use, reasoning and knowledge and writing.
Head to head
Every fact side by side
| Qwen3-8B | Claude Sonnet 4.5 | |
|---|---|---|
| Price per million tokens | $0.12 in · $0.46 out | $3.00 in · $15.00 out |
| Blended (3 in : 1 out) | $0.20 | $6.00 |
| Cheapest host, blended | – | $6.00 (Amazon Bedrock) |
| Intelligence Index | 7.3 | 20.7 |
| Spring Prompt overall | – | – |
| Output speed, tokens a second | – | – |
| Context window | 131,072 tokens | 1,000,000 tokens |
| Released | 28 Apr 2025 | 29 Sep 2025 (newer) |
| Weights | Open | 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 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.
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.
| Measure | Qwen3-8B | Claude Sonnet 4.5 |
|---|---|---|
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 212 · 102 of 244 measured | 7.3thinking reasoning | 20.7thinking reasoning |
| GPQA Diamond% of questions, higher is better 183 · 94 of 231 measured | 58.9%thinking reasoning | 83.4%thinking reasoning |
| Humanity's Last Exam% of questions, higher is better 221 · 117 of 243 measured | 3.9%thinking reasoning | 17.8%thinking reasoning |
| Berkeley Function Calling Leaderboard (BFCL) V4 reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 3 results | ||
| Overall accuracy · BFCL% correct, higher is better 30 · 2 of 81 measured | 42.6%bfcl fc | 73.2%bfcl fc |
| Irrelevance detection · BFCL% correct, higher is better 38 · 5 of 81 measured | 82.3%bfcl prompt | 95.0%bfcl prompt |
| Memory · BFCL% correct, higher is better 40 · 2 of 81 measured | 14.6%bfcl fc | 65.0%bfcl fc |
| UGI Leaderboard reported by UGI Leaderboard · 3 results | ||
| Writing score · UGIscore out of 100, higher is better 162 · 23 of 230 measured | 28.0no reasoning | 66.9no reasoning |
| Requested-length error · UGI% off the requested word count, lower is better 83 · 15 of 230 measured | 17.0%no reasoning | 5.0%no reasoning |
| Style adherence · UGIscore from 0 to 1, higher is better 60 · 8 of 230 measured | 0.36no reasoning | 0.41thinking reasoning |
| Vectara Hallucination Leaderboard reported by Vectara Hallucination Leaderboard · 2 results | ||
| Hallucination rate · Vectara% of summaries, lower is better 9 · 77 of 104 measured | 4.8% | 12.0% |
| Answer rate · Vectara% of documents, higher is better 17 · 86 of 104 measured | 99.9% | 95.6% |
Show all 25 results (14 more)
| Measure | Qwen3-8B | Claude Sonnet 4.5 |
|---|---|---|
| Artificial Analysis reported by Artificial Analysis · 9 results | ||
| Terminal-Bench 2.1% of tasks, higher is better 122 · 53 of 130 measured | 2.2%thinking reasoning | 55.8%thinking reasoning |
| Terminal-Bench 4.0% of tasks, higher is better 67 · 67 of 113 measured | 0.0%thinking reasoning | 0.0%thinking reasoning |
| τ-bench banking% of tasks, higher is better 109 · 40 of 121 measured | 4.7%thinking reasoning | 24.5%thinking reasoning |
| Artificial Analysis Coding Indexindex score, higher is better 123 · 55 of 131 measured | 9.0thinking reasoning | 52.1thinking reasoning |
| IFBench% of instructions, higher is better 165 · 85 of 195 measured | 33.5%thinking reasoning | 57.3%thinking reasoning |
| Long-context reasoning (AA-LCR)% of questions, higher is better 223 · 97 of 234 measured | 0.0%no reasoning | 72.3%thinking reasoning |
| SciCode% of problems, higher is better 105 · 63 of 114 measured | 27.9%thinking reasoning | 45.7%thinking reasoning |
| Terminal-Bench Hard% of tasks, higher is better 173 · 47 of 190 measured | 2.3%no reasoning | 35.6%thinking reasoning |
| τ²-bench telecom% of tasks, higher is better 142 · 73 of 193 measured | 27.8%thinking reasoning | 78.1%thinking reasoning |
| Berkeley Function Calling Leaderboard (BFCL) V4 reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 5 results | ||
| Multi-turn tasks · BFCL% correct, higher is better 23 · 9 of 81 measured | 41.8%bfcl fc | 61.4%bfcl fc |
| Relevance detection · BFCL% correct, higher is better 8 · 63 of 81 measured | 93.8%bfcl fc | 68.8%bfcl fc |
| Single-turn calls (curated) · BFCL% correct, higher is better 14 · 12 of 81 measured | 88.6%bfcl prompt | 88.7%bfcl fc |
| Single-turn calls (user-contributed) · BFCL% correct, higher is better 9 · 5 of 81 measured | 80.5%bfcl fc | 81.1%bfcl fc |
| Web search · BFCL% correct, higher is better 35 · 6 of 81 measured | 13.5%bfcl prompt | 81.0%bfcl fc |
Compare them with other models
Quick answers
Is Qwen3-8B better than Claude Sonnet 4.5?
Pick Claude Sonnet 4.5 if quality matters most: it scores higher (Intelligence Index 20.7 against 7.3). Pick Qwen3-8B if cost matters more: it costs 30 times less per token.
Which is cheaper, Qwen3-8B or Claude Sonnet 4.5?
Qwen3-8B is cheaper: $0.20 against $6.00 per million tokens, 30 times less.
Which is better for agents and tool use, Qwen3-8B or Claude Sonnet 4.5?
Claude Sonnet 4.5 ranks 102 of 166 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 130 for Qwen3-8B.
Which is better for reasoning and knowledge, Qwen3-8B or Claude Sonnet 4.5?
Claude Sonnet 4.5 ranks 100 of 200 models on sale for hard questions across science, maths and general knowledge, against 171 for Qwen3-8B.
Which is better for writing, Qwen3-8B or Claude Sonnet 4.5?
Claude Sonnet 4.5 ranks 35 of 189 models on sale for what people prefer in blind comparisons, and judged writing quality, against 160 for Qwen3-8B.
Which is better for sticking to the facts, Qwen3-8B or Claude Sonnet 4.5?
Qwen3-8B ranks 7 of 153 models on sale for summarising without inventing things, and factual answers, against 102 for Claude Sonnet 4.5.
Which has the bigger context window, Qwen3-8B or Claude Sonnet 4.5?
Claude Sonnet 4.5: 1,000,000 tokens, against 131,072 for Qwen3-8B.
Run Qwen3-8B and Claude Sonnet 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.