vs
2:15
overall capability wins (1 ties across 18 scored). Qwen3.8 Max leads overall.
- Reasoning1:5
- Coding0:6
- Arena0:1
- Agents0:1
- Tool use1:0
- Composite indices0:2
- Specs0:2
Verdict
Qwen3.8 Max leads coding (widest gap: +77.3 on WebDev Elo); Qwen3.8 Max ranks higher on LMArena (+44 Elo); Qwen3.8 Max offers ~3.8x larger context (1M vs 262K).
- Qwen3.8 Max leads coding (widest gap: +77.3 on WebDev Elo)
- Qwen3.8 Max ranks higher on LMArena (+44 Elo)
- Qwen3.8 Max offers ~3.8x larger context (1M vs 262K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Qwen3.8 27B
- -
- Qwen3.8 Max
- $3.50
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
Aug 14, 2026
Aug 3, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
Alibaba Cloud
Modalities
text, image, video → text
text, image, video → text
Specs
Context window
262K
1M
Max output
131K
131K
Parameters
27B
2.4T (95B act.)
Pricing
Input $/1M
—
$2
Output $/1M
—
$6
Blended $/1M (3∶1)
-
$3
Speed
tok/s
-
86
TTFT (s)
-
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
90.5
92.8
Humanity's Last Exam
33.9
43.1
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
15.6
31.7
AA-LCR v1.1
82.0
80.3
CritPt
5.4
17.7
MMMU-Pro
76.3
82.8
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
61.7
67.7
SWE-bench Multilingual
—
—
LiveCodeBench
90.3
—
Terminal-Bench 2.1
79.8
81.3
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
46.6
52.1
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
42.3
55.9
DeepSWE
42.2
56.6
WebDev Arena
1,593
1,671
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,437
1,481
| Metric | Qwen3.8 27B | Qwen3.8 Max | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Alibaba | Alibaba | — |
| Family | Qwen | Qwen | — |
| License | Apache 2.0 | Qwen3.8-Max License | — |
| Open weights | Yes | Yes | — |
| Release | Aug 14, 2026 | Aug 3, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | Alibaba Cloud | — |
| Modalities | text, image, video → text | text, image, video → text | — |
| Specs | |||
| Context window | 262K | 1M | B +738K |
| Max output | 131K | 131K | tie |
| Parameters | 27B | 2.4T (95B act.) | B +2K |
| Pricing | |||
| Input $/1M | — | $2 | — |
| Output $/1M | — | $6 | — |
| Blended $/1M (3∶1) | - | $3 | — |
| Speed | |||
| tok/s | - | 86 | — |
| TTFT (s) | - | 0.4 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 90.5 | 92.8 | B +2.3 pts |
| Humanity's Last Exam | 33.9 | 43.1 | B +9.2 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 15.6 | 31.7 | B +16.1 pts |
| AA-LCR v1.1 | 82.0 | 80.3 | A +1.7 pts |
| CritPt | 5.4 | 17.7 | B +12.3 pts |
| MMMU-Pro | 76.3 | 82.8 | B +6.5 pts |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | 61.7 | 67.7 | B +6.0 pts |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 90.3 | — | — |
| Terminal-Bench 2.1 | 79.8 | 81.3 | B +1.5 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 46.6 | 52.1 | B +5.5 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | 42.3 | 55.9 | B +13.6 pts |
| DeepSWE | 42.2 | 56.6 | B +14.4 pts |
| WebDev Arena | 1,593 | 1,671 | B +77 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,437 | 1,481 | B +44 Elo |
Qwen3.8 27B
- Context
- 262K / 131K out
- Parameters
- 27B
- Price
- —
- Speed
- —
- Modalities
- text, image, video → text
- License
- Apache 2.0
Open-weight Qwen3.8 27B dense vision-language model (Apache-2.0, released Aug 14, 2026) — native image and video understanding, 262K context extensible to 1M, flexible thinking control via reasoning_effort, and strong agentic coding for its size (Terminal-Bench 2.1 73.0, SWE-bench Pro 61.7, GPQA Diamond 89.2).
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.