Qwen3 8B
Apr 2025
Qwen3 8BCompare inon
Benchmarks of Qwen3 8B26
BenchmarkByScoreRank
Overall1
Knowledge5
Science3
Coding3
Agents4
Documents2
Legacy · saturated or no longer reported8
No benchmark matches
Rankingin non-thinking mode?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperClaude Haiku 5.5max×7.2-35%$0.21Remove sampling parameters+5
BetterGPT-6 Lunalow×3.6-35%$0.005Function calling only with none · Chat Completions+3IOpenK2 Horizon 7Bthinking×3.4+69%$0.30 / $1.20Memory (4-bit): 4.1 GB → 4.5 GB
CheaperMiniMax M3.1 Flashthinking×7.1-15%$0.22Complete model response must be appended
CheaperGLM 5.3 FlashmaxOpen×7-23%$0.28Memory (4-bit): 4.1 GB → 193 GB+5ModelScoreCostPer taskWhat changes
Claude Haiku 5.5max×7.2-35%$0.21Remove sampling parameters+5MiniMax M3.1 Flashthinking×7.1-15%$0.22Complete model response must be appendedGLM 5.3 FlashmaxOpen×7-23%$0.28Memory (4-bit): 4.1 GB → 193 GB+5Qwen3.8 FlashthinkingOpen×6.6-26%$0.37Memory (4-bit): 4.1 GB → 108 GB+6GPT-6 Lunamax×6.4-35%$0.068Function calling only with none · Chat Completions+3ModelScoreCostPer taskWhat changes
GPT-6 Lunalow×3.6-35%$0.005Function calling only with none · Chat Completions+3
Granite 4.2 3BthinkingOpen+51%-83%$0.006
Gemma 4 E4BthinkingOpen+49%-87%$0.008GPT-5.6 Lunalow×3.5+45%$0.01GPT OSS 20BhighOpen+50%-88%$0.011Memory (4-bit): 4.1 GB → 13 GBModelScoreCostPer taskWhat changes
IK2 Horizon 7BthinkingOpen×3.4+69%$0.30 / $1.20Memory (4-bit): 4.1 GB → 4.5 GBIK2 Horizon 3.7BthinkingOpen×2.6+69%$0.30 / $1.20Gemma 4 12BthinkingOpen×2.4-52%$0.10 / $0.30Memory (4-bit): 4.1 GB → 6 GBSApriel v1.5 15B ThinkerOpen×2.3——Memory (4-bit): 4.1 GB → 7.4 GBSApriel v1.6 15B ThinkerOpen×2.2——Memory (4-bit): 4.1 GB → 7.4 GBModelScoreCostPer taskWhat changes
Qwen3.8 Max (0902)thinking×7.6—$5.41Qwen3.8 2.4T A95BthinkingOpen×6.7—$2.16Memory (4-bit): 4.1 GB → 1,468 GB+5Qwen3.8 FlashthinkingOpen×6.6—$0.37Memory (4-bit): 4.1 GB → 108 GB+6Qwen3.7 Maxthinking×4.9—$1.15Qwen3.6 Max Previewthinking×4.7—$1.30 / $7.80Include the word JSON · QwenCloudQwen3.8 27BmediumOpen×4.6—$1.13Memory (4-bit): 4.1 GB → 17 GB+5Qwen3.6 Plusthinking×4.5—$0.50 / $3Qwen3.7 Plusthinking×4.2—$0.22All Alibaba (Qwen) models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · compressed✓ 16 GB · 8-bit✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use
Benchmarks
all categoriesChat & Knowledge
Humanity's Last Exam2#336
AA Omniscience Index-74#303
CritPt (physics)0#180
AA Omniscience · Accuracy11.1#273
GPQA Diamond45.2#270
AIME 2025 · AA archived24.3#112
Coding
LiveCodeBench · AA archived20.2#135
Agents & Tool Use
τ²-bench · AA24.9#174
Writing & Language
IFBench (instructions)28.6#217
COMET22 averagelab82.7
COMET22 multi-domain Zh-Enlab86.4
COMET22 multi-domain En-Zhlab82
COMET22 multi-domain En-Delab79.8
COMET22 multi-domain De-Enlab82.7
COMET22 WMT24lab80.8
Acceptance averagelab91
Acceptance En-Zhlab88
Acceptance Zh-Enlab97
Acceptance Zh-Xlab90.5
Acceptance X-Zhlab87.1
Excellence averagelab72
Excellence En-Zhlab68
Excellence Zh-Enlab76
Excellence Zh-Xlab72
Excellence X-Zhlab71.5
FAQ
How much does Qwen3 8B cost?›
$0.18 per 1M input tokens and $0.70 per 1M output tokens.
Is Qwen3 8B free to use?›
There is no free API; the cheapest paid price is $0.18 per 1M input tokens. The weights are open, so you can run it yourself.
How good is Qwen3 8B, and what is it best at?›
#333 of 334 in Chat & Knowledge, #75 of 132 in Coding, #125 of 133 in Agents & Tool Use, #214 of 245 in Writing & Language, #112 of 133 in Vision & Documents.
Is Qwen3 8B better than Hermes 3 70B Instruct?›
On AA Intelligence Index 4.3.2, Qwen3 8B scores 6 vs 6 — about the same.
How can I use Qwen3 8B?›
Through the API of Alibaba.
What is the context window of Qwen3 8B?›
33K tokens, with up to 8K output tokens.
When was Qwen3 8B released? Is there a newer version?›
Released 2025-04-27. It is the current version in its line.
Is Qwen3 8B open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run Qwen3 8B locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Qwen3 8B read images?›
No, it accepts text only.
What is Qwen3 8B?›
Qwen3 8B is an open-weights language model from Alibaba (Qwen), released Apr 2025.
What are the best alternatives to Qwen3 8B?›
Claude Haiku 5.5 (cheaper), GPT-6 Luna (better), K2 Horizon 7B (open weights).