MAI-Thinking-1 vs Qwen3.5 397B

Change models
IndexAll categories, out of 100.
53.3 ±8.1
53.6 ±2.4 best
GapIndex points below the top.
−0.3 · no clear gap
Top
CoverageIndex weight that has results.
70%
100% best
SpeedOutput tokens per second.
—
54/s
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2026-06
2026-02
Input / 1MUS dollars per 1M input tokens.
—
$0.60
Output / 1MUS dollars per 1M output tokens.
—
$3.60
1K chats2K tokens in, 500 out, each.
—
$3.00
1K documents20K tokens in, 1K out, each.
—
$15.60
1K agent tasks200K tokens in, 20K out, each.
—
$192.00
ContextMaximum tokens in one request.
256K best
128K
AgenticMulti-step tasks with tools.
46.5
53.0 best
CodingCode writing and repair.
55.0 best
53.0
ReasoningLogic problems and puzzles.
—
51.7
MultimodalTasks with images and text.
—
55.3
KnowledgeFacts and expert knowledge.
52.8
53.6 best
MultilingualTasks in many languages.
—
55.4
InstructionTasks with strict prompt rules.
68.2 best
55.4
MathMath problems.
52.3
53.4 best

Benchmarks

9 shared
Only MAI-Thinking-1: 3Only Qwen3.5 397B: 60
MAI-Thinking-1
Qwen3.5 397B
84.2%55 / 141
88.4% best36 / 141
73.5%52 / 114
76.2% best42 / 114
52.8% best62 / 83
50.9%69 / 83
85.0%20 / 75
87.8% best5 / 75
87.7% best15 / 59
83.6%25 / 59
85.0% best2 / 57
76.5%25 / 57
46.0%34 / 41
52.5% best26 / 41
94.5% best19 / 34
93.3%21 / 34
84.9%17 / 27
87.9% best11 / 27