AlibabaXhigh Released August 2026
How Qwen3.8 27B compares with the other 88 ranked language models.
Better for simple tasks. Ranked #66 of 89 on the overall intelligence score, ahead of 26% of models.
Slow writer. Writes about 29 words per second, faster than 15% of models.
Thinks before answering. You wait about 52 sec before the answer starts, because it reasons through the problem first.
Mid-range price. $1.13 per million tokens, cheaper than 65% of models.
Good for
Not ideal for
Intelligence
20
overall score
#66 of 89typical 30
Coding
–
coding score
typical –
Speed
39
tokens per second
#76 of 89typical 91
Wait
52 sec
before it replies
#65 of 79typical 15s
Price
$1.13
per 1M tokens
#30 of 84typical $2.00
Scores on individual tests that make up the intelligence score. The faint line marks the best score of any model.
Humanity's Last Exam
Very hard expert questions across many subjects
GPQA Diamond
PhD-level biology, physics and chemistry questions
SciCode
Writing code to solve science research problems
Long documents
Finding and reasoning over facts in very long texts
Terminal-Bench
Doing real tasks on a computer by itself, like an agent
Customer support
Helping bank customers as a support agent using tools
This model next to the models closest to it. Click another bar to open that model.
Through the API, at the listed price. Chat apps usually charge a monthly plan instead.
1,000 chat replies
a question and a ~350-word answer, 1,000 times
Summarise a 100-page report
about 50,000 words in, a one-page summary out
Write a 2,000-word article
a short brief in, a long article out
Input, per 1M tokens
$0.50
Output, per 1M tokens
$3.00
Median times measured through the API.
Most of the wait is the model thinking before it writes. That makes answers better on hard questions, but slow for quick chats.
Higher effort settings think longer: smarter, but slower and pricier.
Other companies' models with the closest intelligence score.
Benchmark data from Artificial Analysis, an independent benchmarking company. Rankings refresh every few hours.
kitio.ai