text model comparison
Qwen3.8-Max vs Qwen4-Max
Qwen3.8-Max and Qwen4-Max are both current text models, from Alibaba Qwen. Qwen4-Max is 25% cheaper per 1M input tokens than Qwen3.8-Max, $1.50 vs $2. On Arena WebDev, Qwen3.8-Max scores 1670 Elo and Qwen4-Max 1750 Elo (Arena, measured November 15, 2026). Both accept 1,000,000 tokens of context.
- Higher Arena WebDev
- Qwen4-Max1750 Elo against 1670 Elo
- Lower price per 1M input tokens
- Qwen4-Max$1.50 against $2
- Newer release
- Qwen4-MaxNov 2026 against Aug 2026
Data verified November 17, 2026 · Dataset 2026-11-17
Side by side
Qwen3.8-Max and Qwen4-Max, line by line
Prices in effect today, compared at the same tier wherever both list one. The better value in each row is highlighted.
| Fact | Qwen3.8-Max | Qwen4-Max |
|---|---|---|
| Price | ||
| Price per 1M input tokens | $2 | $1.50 (better) |
| Price per 1M output tokens | $6 | $4.50 (better) |
| Independent benchmarks | ||
| Arena WebDev | 1670 Elo (#13 of 46)Range 1658 to 1683 | 1750 Elo (#6 of 46) (better)Range 1734 to 1765 |
| Arena Text | 1484 Elo (#12 of 50)Range 1480 to 1489 | Not measured |
| Epoch Capabilities Index | 157.6 points (#22 of 43)Range 155.4 to 160 | Not measured |
| GPQA Diamond | 92.7% (#20 of 40) | Not measured |
| FrontierMath Tiers 1-3 (v2) | 74.7% (#19 of 39) | Not measured |
| SimpleQA Verified | 45.8% (#23 of 36) | Not measured |
| Specs | ||
| Provider | Alibaba Qwen | Alibaba Qwen |
| Release date | August 2, 2026 | November 16, 2026 |
| Access | API only | API only |
| Context window | 1,000,000 tokens | 1,000,000 tokens |
| Max output | 131,072 tokens | 131,072 tokens |
| Input | text, image, video | text, image, video |
| Output | text | text |
Ranges under the scores are the publishers' own: published rating ranges from Arena and 90% credible intervals from Epoch AI. Where two ranges overlap, the gap is inside the measurement's own uncertainty, so the row says too close to call and neither score is highlighted or counted as a lead.
Provider results
What the labs report
Results tables a lab published with one of its models, where it lists both Qwen3.8-Max and Qwen4-Max. These are the lab's own figures, not independent measurements, and each table is read on its own: labs run benchmarks with different harnesses and budgets.
Qwen Team, Qwen4-Max launch chart, November 16, 2026Qwen Team · 2026-11-17
| Benchmark | Qwen3.8-Max | Qwen4-Max |
|---|---|---|
| DeepSWE v1.1 | 69.3% | 75.7% (better) |
| Terminal-Bench 4.0 | 38.9% | 48.5% (better) |
| ProgramBench | 28.0% | 31.5% (better) |
| Agents' Last Exam v1.1 | 27.9% | 46.7% (better) |
| APEX-Agents v1.1 | 63.3% | 72.1% (better) |
| HealthBench Professional | 39.6% | 60.0% (better) |
| Humanity's Last Exam | 43.9% | 51.7% (better) |
| BioMysteryBench | 84.1% | 85.0% (better) |
| Chartography | 42.0% | 54.2% (better) |
Alibaba's chart from the Qwen4-Max announcement. Qwen3.8-Max is the 0902 refresh, and Kimi K3.1 is Kimi K3.1 Pro: its DeepSWE and BioMysteryBench figures match Moonshot's own card. The APEX-Agents and Agents' Last Exam scores come from those benchmarks' leaderboards; Alibaba ran Chartography for its own two models with a larger image budget. Left out here: CodeArena and AgentArena (Arena ratings), AA-Briefcase and AutomationBench-AA (Artificial Analysis), HLAB-Vals-1.1 (taken from the Vals Index), and JobBench and Video-Index, which we could not confirm are public.
Charts
Against every current text model
Where Qwen3.8-Max and Qwen4-Max land among the current models of their task, on every independent benchmark, on price and on context.
Where each stands
On Arena WebDev, Qwen3.8-Max ranks #13 and Qwen4-Max #6 of 46 current models. Each bar is the share of current models it beats: on score, price and context.
Independent benchmarks
Arena Text
Qwen3.8-Max:#12of 50
1,484 Elo · ahead of 78%
Published rating range 1,480 to 1,489
Qwen4-Max: Not scored
Arena WebDev
Qwen3.8-Max:#13of 46
1,670 Elo · ahead of 73%
Published rating range 1,658 to 1,683
Qwen4-Max:#6of 46
1,750 Elo · ahead of 89%
Published rating range 1,734 to 1,765
Epoch Capabilities Index
Qwen3.8-Max:#22of 43
157.6 · ahead of 50%
90% credible interval 155.4 to 160
Qwen4-Max: Not scored
GPQA Diamond
Qwen3.8-Max:#20of 40
92.7% · ahead of 49%
Qwen4-Max: Not scored
FrontierMath Tiers 1-3 (v2)
Qwen3.8-Max:#19of 39
74.7% · ahead of 53%
Qwen4-Max: Not scored
Price and context
Price per 1M input tokens
Qwen3.8-Max:#36of 60
$2 · cheaper than 22%
Qwen4-Max:#33of 60
$1.50 · cheaper than 42%
Price per 1M output tokens
Qwen3.8-Max:#34of 60
$6 · cheaper than 39%
Qwen4-Max:#32of 60
$4.50 · cheaper than 47%
Context window
Qwen3.8-Max:#33of 71
1M tokens · longer than 21%
Qwen4-Max:#33of 71
1M tokens · longer than 21%
Quality against price
Qwen3.8-Max scores 1,484 at $3 (blended price per 1M tokens). The line joins the current models no other beats on both price and Arena Text score; hover any point to name it.
Quality against price: the data
| Model | Provider | Blended price per 1M tokens | Arena Text (Elo) |
|---|---|---|---|
| Gemini 4 Carbon | $4 | 1,526 | |
| Gemini 4 Argon | $4 | 1,522 | |
| Muse Wonder | Meta | $8 | 1,511 |
| Claude Opus 5.5 | Anthropic | $8 | 1,507 |
| Gemini 3.8 Flash | $1.50 | 1,497 | |
| Muse Spark 1.3 | Meta | $2 | 1,497 |
| Gemini 4 Hydrogen | $3 | 1,490 | |
| Kimi K3 | Moonshot Kimi | $6 | 1,487 |
| GPT-5.6 Sol | OpenAI | $8 | 1,485 |
| Kimi K3.1 Pro | Moonshot Kimi | $4 | 1,485 |
| Qwen3.8-Max | Alibaba Qwen | $3 | 1,484 |
| Gemini 3.6 Flash | $3 | 1,483 | |
| GLM-5.3 | Z.ai | $2.15 | 1,482 |
| GPT-5.5 | OpenAI | $11 | 1,481 |
| GPT-6.1 Sol | OpenAI | $4 | 1,481 |
| MiMo-V2.6-Pro | Xiaomi MiMo | $0.54 | 1,480 |
| GPT-5.4 | OpenAI | $5.63 | 1,475 |
| DeepSeek-V4.1-Flash | DeepSeek | $0.52 | 1,475 |
| GLM-5.3-Flash | Z.ai | $0.24 | 1,475 |
| GPT-6 Astra | OpenAI | $20 | 1,474 |
| Gemini 3 Flash Preview | $1.13 | 1,473 | |
| Claude Sonnet 5.5 | Anthropic | $4 | 1,473 |
| Grok 4.20 | xAI | $1.56 | 1,472 |
| GPT-5.6 Terra | OpenAI | $4.50 | 1,468 |
| Grok 4.5 | xAI | $3 | 1,466 |
| DeepSeek-V4-Pro | DeepSeek | $1.98 | 1,464 |
| Grok 4.6 | xAI | $3 | 1,457 |
| GPT-6 Sol | OpenAI | $4 | 1,457 |
| Gemini 3.5 Flash-Lite | $0.85 | 1,455 | |
| GPT-5.6 Luna | OpenAI | $0.45 | 1,453 |
| Step 5 Preview | StepFun | $1.43 | 1,449 |
| MiMo-V2.6-Flash | Xiaomi MiMo | $0.18 | 1,448 |
| Grok 4.7 | xAI | $3 | 1,443 |
| Grok 4.3 | xAI | $1.56 | 1,442 |
| GPT-6 Luna | OpenAI | $0.20 | 1,441 |
| Claude Haiku 5.5 | Anthropic | $0.20 | 1,421 |
| Mistral Large 4 | Mistral AI | $1.05 | 1,420 |
| GPT-4.1 | OpenAI | $3.50 | 1,415 |
| Inkling 2 Small | Thinking Machines Lab | $0.47 | 1,412 |
| Solar Pro 4 | Upstage | $0.52 | 1,387 |
| GPT-4.1 mini | OpenAI | $0.70 | 1,383 |
| GPT-4.1 nano | OpenAI | $0.18 | 1,322 |
Cost
What each costs
Example workloads at each provider's listed rates, then every other rate either one lists. The cheaper figure in each row is highlighted.
The same work, priced
For 10M input and 2M output tokens, Qwen3.8-Max costs 1.3 times as much as Qwen4-Max, at each provider's listed rates.
10M input and 2M output tokens
Qwen3.8-Max costs 1.3× as much
- Qwen3.8-Max
- $32
- Qwen4-Max
- $24
| Workload or rate | Qwen3.8-Max | Qwen4-Max |
|---|---|---|
| Example workloads | ||
| 10M input and 2M output tokens | $32 | $24 (better) |
| The same, with 9M of the input tokens cached | Not listed | $11.18 |
| Other listed rates | ||
| 1M input tokens, cached | Not listed | $0.075 |
| 1M input tokens, 5-minute cache write | Not listed | $1.875 |
| 1M input tokens, 1-hour cache write | Not listed | $3 |
Your own bill will differ. These figures use list prices and example workloads. The AI Model Cost Audit reviews your real usage, shortlists models that score as well for each task for less, and estimates the savings.
Launch price $100, regular price $150
Verdict
When to choose each
Qwen3.8-Max
Where it leads
No measured lead over Qwen4-Max in this data.
Best for
- Workloads already tuned on Qwen3.8-Max; new work can start on the cheaper Qwen4-Max
- Agentic and coding evaluations that need a 1M-token window at mid-tier cost
Watch out for
- Two releases share the name: the original 2 August 2026 model and the 0902 refresh, a separate id at the same price
- The recorded rates are the international (Singapore) price; the Global and Beijing listings charge $1.65 input and $4.951 output per 1M tokens
- Alibaba publishes no fixed cached-input rate for it
Qwen4-Max
Where it leads
- Lower input and output prices, $1.50 and $4.50 vs $2 and $6 per 1M tokens
- Higher Arena WebDev score, 1750 Elo vs 1670 Elo
- A cached-input rate, $0.075 per 1M tokens
Best for
- Coding and long-horizon agent work at a lower price than Claude Opus 5.5 or GPT-6.1 Sol
- Office deliverables and research over long documents, images and video in one 1M-token context
Watch out for
- API only: Alibaba has not released the weights, and says only that smaller Qwen4 models are coming to open source
- Artificial Analysis scores it 43 on its Intelligence Index against 59 for Claude Opus 5.5 (November 2026), and Alibaba's own chart has Opus 5.5 ahead on Terminal-Bench 4.0 (59.6% against 48.5%) and ProgramBench (65.0% against 31.5%), with Kimi K3.1 Pro at 67.2% on Terminal-Bench
- The recorded rates are the Singapore (international) price; Beijing and the Global regions charge $1.413 input and $4.24 output per 1M tokens, and there is no batch discount
- Arena Text has not rated it yet, and its WebDev rating rests on 1,814 votes so far
From benchmark to production
Building on Qwen3.8-Max or Qwen4-Max?
Public scores rarely hold on your own data. We test both models on your real tasks, then build the agent or pipeline around the one that wins.
AI Model Cost AuditLaunch price $100, regular price $150
FAQ
Qwen3.8-Max vs Qwen4-Max FAQ
Which is cheaper, Qwen3.8-Max or Qwen4-Max?
Qwen4-Max is 25% cheaper per 1M input tokens than Qwen3.8-Max, $1.50 vs $2. For 10M input and 2M output tokens, Qwen3.8-Max costs $32 and Qwen4-Max $24 at standard rates.
Do Qwen3.8-Max and Qwen4-Max discount cached input?
Qwen4-Max lists a cached-input rate of $0.075 per 1M tokens, 95% below its standard input rate. No cached-input rate is listed for Qwen3.8-Max.
Which scores higher on independent benchmarks, Qwen3.8-Max or Qwen4-Max?
Qwen4-Max leads on Arena WebDev, 1750 Elo to 1670 Elo. Scores are independent evaluations, not either vendor's launch figures.
How do Qwen3.8-Max and Qwen4-Max compare in Qwen Team's own results?
In the Qwen4-Max launch chart (November 16, 2026), Qwen4-Max leads on DeepSWE v1.1, 75.7% to 69.3%; Qwen4-Max leads on Terminal-Bench 4.0, 48.5% to 38.9%; Qwen4-Max leads on ProgramBench, 31.5% to 28.0%. These are Qwen Team's figures, not independent measurements.
Which has the larger context window, Qwen3.8-Max or Qwen4-Max?
Both accept 1,000,000 tokens of context. Both return up to 131,072 tokens per response.
AI Model Brief
The AI models worth knowing, once a week
What shipped, how it scores on independent benchmarks, what it costs per million tokens and what was deprecated, from the data behind our AI model benchmarks.
We use your address only to send the brief. See our privacy policy.
See this week’s briefFurther reading
More on Qwen3.8-Max and Qwen4-Max
Related comparisons
- Gemini 4 Argon vs Qwen3.8-Max
- Claude Opus 5.5 vs Qwen4-Max
- Claude Sonnet 5.5 vs Qwen3.8-Max
- GPT-6.1 Sol vs Qwen4-Max
- GPT-6 Luna vs Qwen3.8-Max
- Kimi K3.1 Pro vs Qwen4-Max
- DeepSeek-V4.1-Flash vs Qwen3.8-Max
- Gemini 3.8 Flash vs Qwen3.8-Max
- GLM-5.3 vs Qwen3.8-Max
- DeepSeek-V4-Pro vs Qwen3.8-Max
- Claude Sonnet 5 vs Qwen3.8-Max
- Kimi K3 vs Qwen3.8-Max
Production evaluation
Choosing between Qwen3.8-Max and Qwen4-Max?
Public benchmarks narrow the shortlist. We run both on your own tasks, data and cost limits, then build the routing and review controls to run the winner in production.
- A head-to-head evaluation on your real workload
- Routing between models by cost and difficulty
