vs
0:15
overall capability wins (0 ties across 15 scored). Qwen3.8 Max leads overall.
- Reasoning0:6
- Coding0:3
- Arena0:1
- Tool use0:2
- Composite indices0:2
- Pricing3:0
- Speed0:1
- Specs0:3
Verdict
Qwen3.8 Max leads coding (widest gap: +271.7 on WebDev Elo); Qwen3.8 Max ranks higher on LMArena (+39 Elo); Qwen3.8 Max offers ~3.8x larger context (1M vs 262K).
- Qwen3.8 Max leads coding (widest gap: +271.7 on WebDev Elo)
- Qwen3.8 Max ranks higher on LMArena (+39 Elo)
- Qwen3.8 Max offers ~3.8x larger context (1M vs 262K)
- Qwen3.8 Max allows ~4.0x more max output (131K vs 33K)
- Qwen3.5 397B-A17B is ~5.0x cheaper on a blended token basis than Qwen3.8 Max
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Qwen3.5 397B-A17B
- $0.70
- Qwen3.8 Max
- $3.50
Qwen3.5 397B-A17B is about 5.0x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Alibaba
Alibaba
Family
Qwen
Qwen
License
Apache 2.0
Qwen3.8-Max License
Open weights
Yes
Yes
Release
Mar 1, 2026
Aug 3, 2026
Knowledge cutoff
-
-
API / provider
Together / Fireworks (ref.)
Alibaba Cloud
Modalities
text → text
text, image, video → text
Specs
Context window
262K
1M
Max output
33K
131K
Parameters
397B (17B act.)
2.4T (95B act.)
Pricing
Input $/1M
$0.40
$2
Output $/1M
$1.20
$6
Blended $/1M (3∶1)
$0.60
$3
Speed
tok/s
75
86
TTFT (s)
0.4
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
89.3
92.8
Humanity's Last Exam
29.0
43.1
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
30.8
31.7
AA-LCR v1.1
77.3
80.3
CritPt
1.7
17.7
MMMU-Pro
77.3
82.8
IFBench
78.8
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
67.7
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
51.3
81.3
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
44.8
52.1
CursorBench
—
—
SWE-Rebench
59.9
—
NL2Repo-Bench
—
55.9
DeepSWE
—
56.6
WebDev Arena
1,399
1,671
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,442
1,481
| Metric | Qwen3.5 397B-A17B | Qwen3.8 Max | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Alibaba | Alibaba | — |
| Family | Qwen | Qwen | — |
| License | Apache 2.0 | Qwen3.8-Max License | — |
| Open weights | Yes | Yes | — |
| Release | Mar 1, 2026 | Aug 3, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Together / Fireworks (ref.) | Alibaba Cloud | — |
| Modalities | text → text | text, image, video → text | — |
| Specs | |||
| Context window | 262K | 1M | B +738K |
| Max output | 33K | 131K | B +98K |
| Parameters | 397B (17B act.) | 2.4T (95B act.) | B +2K |
| Pricing | |||
| Input $/1M | $0.40 | $2 | A +$1.60 |
| Output $/1M | $1.20 | $6 | A +$4.80 |
| Blended $/1M (3∶1) | $0.60 | $3 | A +$2.40 |
| Speed | |||
| tok/s | 75 | 86 | B +11 |
| TTFT (s) | 0.4 | 0.4 | tie |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 89.3 | 92.8 | B +3.5 pts |
| Humanity's Last Exam | 29.0 | 43.1 | B +14.1 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 30.8 | 31.7 | B +0.9 pts |
| AA-LCR v1.1 | 77.3 | 80.3 | B +3.0 pts |
| CritPt | 1.7 | 17.7 | B +16.0 pts |
| MMMU-Pro | 77.3 | 82.8 | B +5.5 pts |
| IFBench | 78.8 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | 67.7 | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 51.3 | 81.3 | B +30.0 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 44.8 | 52.1 | B +7.3 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 59.9 | — | — |
| NL2Repo-Bench | — | 55.9 | — |
| DeepSWE | — | 56.6 | — |
| WebDev Arena | 1,399 | 1,671 | B +272 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,442 | 1,481 | B +39 Elo |
Qwen3.5 397B-A17B
- Context
- 262K / 33K out
- Parameters
- 397B (17B act.)
- Price
- $0.40 / $1.20
- Speed
- 75 tok/s · 0.4s TTFT
- Modalities
- text → text
- License
- Apache 2.0
Large open MoE from the Qwen3.5 line under Apache 2.0.
Qwen3.8 Max
- Context
- 1M / 131K out
- Parameters
- 2.4T (95B act.)
- Price
- $2 / $6
- Speed
- 86 tok/s · 0.4s TTFT
- Modalities
- text, image, video → text
- License
- Qwen3.8-Max License
Alibaba's Aug 2026 Max flagship (2.4T MoE / 95B active) — frontier agentic coding via QwenCloud / DashScope with 1M multimodal context and $2/$6 API pricing. Open weights released Aug 12, 2026 as Qwen3.8-2.4T-A95B (text-only, thinking-mandatory, Qwen3.8-Max license); the hosted version adds vision input, non-thinking mode, and built-in tools.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.