Skip to main content
Radar

Every change to the models you run, with its source and its date. Releases, price changes, retirements, API changes, and incidents.Every change to the models you run, with its source.

Follow model changes
BenchLM

Qwen3.5 397B vs Qwen3.8 Max

Decision reading

Qwen3.8 Max has the higher public score, 71.76 versus 54.87, and the 90% score intervals do not overlap.

10 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

Alibaba logo
Model A
Qwen3.5 397B

Alibaba

54.87/100

Estimated · Public rank #99

90% interval 43.4–66.4

Alibaba logo
Model B
Qwen3.8 Max

Alibaba

71.76/100

Supported · Public rank #11

90% interval 68.1–75.4

Updated September 15, 2026. Rank says Qwen3.8 Max is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.

Share or export

Share on XLinkedInSocial cardCSVJSON

Which one for your work

Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.

  • Long documents

    Prompts that approach the documented context limit

    Qwen3.8 Max

    Qwen3.8 Max has the larger documented context window.

    Confidence: documented

Show secondary and unsupported calls

  • Coding work

    Code generation, repair, and software-engineering tasks

    Not enough matched evidence

    Qwen3.5 397B is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.

    Confidence: limited

  • Agentic work

    Tool use, computer use, and multi-step task completion

    Not enough matched evidence

    Qwen3.5 397B is scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.

    Confidence: limited

  • Chat turn cost

    1K fresh input + 500 output tokens

    Not enough matched evidence

    A complete comparable API-rate estimate is not available for both models.

    Confidence: listed-rates

  • Cache-heavy agent loop cost

    200K cached + 20K fresh input + 10K output tokens

    Not enough matched evidence

    The page does not recommend a cost winner because at least one model cannot fit the stated workload in one request. Qwen3.5 397B does not fit this workload in one request. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.8 Max has no comparable published API token rate.

    Confidence: rate-fallback

  • Repository review cost

    50K fresh input + 3K output tokens

    Not enough matched evidence

    A complete comparable API-rate estimate is not available for both models.

    Confidence: listed-rates

What is actually comparable

Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.

Shared results
10
Qwen3.5 397B only
28
Qwen3.8 Max only
50
Like-for-like categories
1 / 8

4 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.

Category results, on a stated basis

Each row shows the public-lane category score for both models: the BenchAlign lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.

Multimodal

Like-for-like
Qwen3.5 397B
63.0
#27/48
Qwen3.8 Max
87.4
#5/48
Basis
Provisional lane · 2 vs 2 weighted rows
Reading
Qwen3.8 Max leads

Agentic

Directional only
Qwen3.5 397B
46.8
Estimated · #74/153
Qwen3.8 Max
67.3
Supported · #9/153
Basis
BenchAlign lane · 13 vs 15 public rows
Reading
Directional only

Coding

Directional only
Qwen3.5 397B
50.4
Estimated · #60/152
Qwen3.8 Max
60.8
Supported · #20/152
Basis
BenchAlign lane · 3 vs 12 public rows
Reading
Directional only

Knowledge

Directional only
Qwen3.5 397B
50.2
Estimated · #82/183
Qwen3.8 Max
68.8
Supported · #17/183
Basis
BenchAlign lane · 6 vs 6 public rows
Reading
Directional only

Instruction following

Directional only
Qwen3.5 397B
0.0
#123/123
Qwen3.8 Max
90.7
#18/123
Basis
Provisional lane · 0 vs 1 weighted rows
Reading
Directional only

Reasoning

Not comparable
Qwen3.5 397B
59.8
Unranked · 2 rankable rows
Qwen3.8 Max
86.5
Unranked · 2 rankable rows
Basis
Provisional lane · 1 vs 2 weighted rows
Reading
Not comparable

Math

Not comparable
Qwen3.5 397B
74.2
Unranked · 5 rankable rows
Qwen3.8 Max
Not ranked
Basis
Provisional lane · 2 vs 0 weighted rows
Reading
Not comparable

Multilingual

Not comparable
Qwen3.5 397B
69.7
#5/12
Qwen3.8 Max
Not ranked
Basis
Provisional lane · 1 vs 0 weighted rows
Reading
Not comparable

Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign) differ from the provisional-lane categories. Unranked scores sit on the lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.

Shape of the matched evidence

Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.

Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.

What each workload costs

Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.

Chat turn

1K fresh input + 500 output tokens

Qwen3.5 397B
$0.0024
Fits in one request
Qwen3.8 Max
API rate not published
Fits in one request

Qwen3.8 Max has no comparable published API token rate.

Repository review

50K fresh input + 3K output tokens

Qwen3.5 397B
$0.0408
Fits in one request
Qwen3.8 Max
API rate not published
Fits in one request

Qwen3.8 Max has no comparable published API token rate.

Cache-heavy agent loop

200K cached + 20K fresh input + 10K output tokens

Qwen3.5 397B
$0.168
Does not fit in one request
Cached input priced at the published list-input rate
Qwen3.8 Max
API rate not published
Fits in one request
Cached-input rate unavailable

Qwen3.5 397B does not fit this workload in one request. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.8 Max has no comparable published API token rate.

Specification differences

Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.

Cached-input rate

A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.

Qwen3.5 397B

Not published

Qwen3.8 Max

No comparable hosted API rate

Alibaba Cloud Model Studio pricing

Documented inputs

Qwen3.5 397B

Not sourced

Qwen3.8 Max

Not sourced

Documented outputs

Qwen3.5 397B

Not sourced

Qwen3.8 Max

Not sourced

Provider availability

Qwen3.5 397B

Not sourced

Qwen3.8 Max

Not sourced

Reasoning profile

Qwen3.5 397B

Non-Reasoning

Qwen3.8 Max

Reasoning

Weight access

Qwen3.5 397B

Open Weight

Qwen3.8 Max

Open Weight

License

Qwen3.5 397B

Open Weight

Qwen3.8 Max

Open Weight

Release date

Qwen3.5 397B

2026-02-16

Qwen3.8 Max

2026-08-03

If you already use one of these models
Deployment change
Both entries list Alibaba as the provider. Confirm endpoint, model ID, limits, and feature support before switching.
Quality signal
Qwen3.8 Max has the higher public score, 71.76 versus 54.87, and the 90% score intervals do not overlap.
Workload cost
A complete comparable API-rate estimate is not available for both models.
Context tradeoff
Qwen3.8 Max has the larger documented window (1M).

Run the same representative tasks against both endpoints before changing production traffic.

Benchmark evidence

The full public result ledger is available for audit without forcing a wide desktop table onto a phone.

Browse raw public benchmark evidence88 rows

Agentic

  • Terminal-Bench 2.0

    Qwen3.5 397B52.5%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • BrowseComp

    Qwen3.5 397B62%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • Claw-Eval

    Qwen3.5 397B56.8%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • QwenClawBench

    Qwen3.5 397B51.8%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • τ³-bench results

    Qwen3.5 397B68.4%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • VITA-Bench

    Qwen3.5 397B43.7%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • DeepPlanning

    Qwen3.5 397B37.6%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • Toolathlon

    Qwen3.5 397B36.3%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MCP Atlas

    Qwen3.5 397B46.1%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MCP-Tasks

    Qwen3.5 397B74.2%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • WideResearch

    Qwen3.5 397B74.0%
    Source
    Qwen3.8 Max81.9%
    Source

    Qwen3.8 Max leads this result

  • Gert Labs

    Qwen3.5 397B46.76%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • ResearchClawBench

    Qwen3.5 397B14.2%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • Terminal-Bench 2.1

    Qwen3.5 397B—
    Qwen3.8 Max86.6%
    Source

    Not directly comparable

  • CoWorkBench

    Qwen3.5 397B—
    Qwen3.8 Max74.8%
    Source

    Not directly comparable

  • JobBench

    Qwen3.5 397B—
    Qwen3.8 Max53.4%
    Source

    Not directly comparable

  • skillsBench

    Qwen3.5 397B—
    Qwen3.8 Max70.2%
    Source

    Not directly comparable

  • Agents' Last Exam

    Qwen3.5 397B—
    Qwen3.8 Max52.4%
    Source

    Not directly comparable

  • AutomationBench

    Qwen3.5 397B—
    Qwen3.8 Max27.3%
    Source

    Not directly comparable

  • Toolathlon-Verified

    Qwen3.5 397B—
    Qwen3.8 Max72.5%
    Source

    Not directly comparable

  • HLE w/ tools

    Qwen3.5 397B—
    Qwen3.8 Max56.2%
    Source

    Not directly comparable

  • OSWorld-Verified

    Qwen3.5 397B—
    Qwen3.8 Max86.1%
    Source

    Not directly comparable

  • OSWorld 2.0

    Qwen3.5 397B—
    Qwen3.8 Max19.4%
    Source

    Not directly comparable

  • WebArena-Verified

    Qwen3.5 397B—
    Qwen3.8 Max66.8%
    Source

    Not directly comparable

  • AndroidWorld

    Qwen3.5 397B—
    Qwen3.8 Max85.3%
    Source

    Not directly comparable

  • MobileWorld

    Qwen3.5 397B—
    Qwen3.8 Max77.8%
    Source

    Not directly comparable

  • Terminal-Bench 2.1 (Vals)

    Qwen3.5 397B—
    Qwen3.8 Max67.4%
    Source

    Not directly comparable

Coding

  • SWE-bench Verified

    Qwen3.5 397B76.2%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • LiveCodeBench v6

    Qwen3.5 397B83.6%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • SWE-bench Pro

    Qwen3.5 397B50.9%
    Source
    Qwen3.8 Max67.7%
    Source

    Qwen3.8 Max leads this result

  • Terminal-Bench 2.1

    Qwen3.5 397B—
    Qwen3.8 Max86.6%
    Source

    Not directly comparable

  • DeepSWE

    Qwen3.5 397B—
    Qwen3.8 Max56.6%
    Source

    Not directly comparable

  • NL2Repo

    Qwen3.5 397B—
    Qwen3.8 Max55.9%
    Source

    Not directly comparable

  • FrontierSWE

    Qwen3.5 397B—
    Qwen3.8 Max73.5%
    Source

    Not directly comparable

  • MLS-Bench Lite

    Qwen3.5 397B—
    Qwen3.8 Max41.0%
    Source

    Not directly comparable

  • PaperBench

    Qwen3.5 397B—
    Qwen3.8 Max93.0%
    Source

    Not directly comparable

  • VulcanBench v3

    Qwen3.5 397B—
    Qwen3.8 Max81.2%
    Source

    Not directly comparable

  • OpenHarmony Bench

    Qwen3.5 397B—
    Qwen3.8 Max60.8%
    Source

    Not directly comparable

  • FrontierSWE v2

    Qwen3.5 397B—
    Qwen3.8 Max15.8%
    Source

    Not directly comparable

  • LiveCodeBench (Vals)

    Qwen3.5 397B—
    Qwen3.8 Max87.9%
    Source

    Not directly comparable

  • SWE-bench (Vals)

    Qwen3.5 397B—
    Qwen3.8 Max85.6%
    Source

    Not directly comparable

Reasoning

  • LongBench v2

    Qwen3.5 397B63.2%
    Source
    Qwen3.8 Max66.3%
    Source

    Qwen3.8 Max leads this result

  • AI-Needle

    Qwen3.5 397B68.7%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MRCRv2

    Qwen3.5 397B—
    Qwen3.8 Max92.9%
    Source

    Not directly comparable

Knowledge

  • GPQA

    Qwen3.5 397B88.4%
    Source
    Qwen3.8 Max92.6%
    Source

    Qwen3.8 Max leads this result

  • SuperGPQA

    Qwen3.5 397B70.4%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MMLU-Pro

    Qwen3.5 397B87.8%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MMLU-Redux

    Qwen3.5 397B94.9%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • C-Eval

    Qwen3.5 397B93%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • HLE

    Qwen3.5 397B28.7%
    Source
    Qwen3.8 Max43.6%
    Source

    Qwen3.8 Max leads this result

  • GPQA-D

    Qwen3.5 397B—
    Qwen3.8 Max92.6%
    Source

    Not directly comparable

  • HLE w/o tools

    Qwen3.5 397B—
    Qwen3.8 Max43.6%
    Source

    Not directly comparable

  • GPQA Diamond (Vals)

    Qwen3.5 397B—
    Qwen3.8 Max93.7%
    Source

    Not directly comparable

  • MMLU-Pro (Vals)

    Qwen3.5 397B—
    Qwen3.8 Max88.6%
    Source

    Not directly comparable

Math

  • AIME26

    Qwen3.5 397B93.3%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • HMMT Feb 2025

    Qwen3.5 397B94.8%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • HMMT Nov 2025

    Qwen3.5 397B92.7%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • HMMT Feb 2026

    Qwen3.5 397B87.9%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MMAnswerBench

    Qwen3.5 397B80.9%
    Source
    Qwen3.8 Max—

    Not directly comparable

Multilingual

  • MMLU-ProX

    Qwen3.5 397B84.7%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • NOVA-63

    Qwen3.5 397B59.1%
    Source
    Qwen3.8 Max—

    Not directly comparable

Multimodal

  • MMMU-Pro

    Qwen3.5 397B79%
    Source
    Qwen3.8 Max82.3%
    Source

    Qwen3.8 Max leads this result

  • MathVision

    Qwen3.5 397B88.6%
    Source
    Qwen3.8 Max95.2%
    Source

    Qwen3.8 Max leads this result

  • CharXiv

    Qwen3.5 397B80.8%
    Source
    Qwen3.8 Max93.5%
    Source

    Qwen3.8 Max leads this result

  • VideoMMMU

    Qwen3.5 397B84.7%
    Source
    Qwen3.8 Max88.7%
    Source

    Qwen3.8 Max leads this result

  • ScreenSpot Pro

    Qwen3.5 397B65.6%
    Source
    Qwen3.8 Max84.5%
    Source

    Qwen3.8 Max leads this result

  • V*

    Qwen3.5 397B95.8%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • MathVision w/ Python

    Qwen3.5 397B—
    Qwen3.8 Max97.7%
    Source

    Not directly comparable

  • BabyVision

    Qwen3.5 397B—
    Qwen3.8 Max82.0%
    Source

    Not directly comparable

  • BabyVision w/ Python

    Qwen3.5 397B—
    Qwen3.8 Max91.3%
    Source

    Not directly comparable

  • ZeroBench

    Qwen3.5 397B—
    Qwen3.8 Max24.0%
    Source

    Not directly comparable

  • ZeroBench w/ Python

    Qwen3.5 397B—
    Qwen3.8 Max49.0%
    Source

    Not directly comparable

  • MedXpertQA (MM)

    Qwen3.5 397B—
    Qwen3.8 Max80.4%
    Source

    Not directly comparable

  • Vision2Web

    Qwen3.5 397B—
    Qwen3.8 Max69.0%
    Source

    Not directly comparable

  • CharXiv w/o tools

    Qwen3.5 397B—
    Qwen3.8 Max88.4%
    Source

    Not directly comparable

  • OmniDocBench 1.5

    Qwen3.5 397B—
    Qwen3.8 Max92.1%
    Source

    Not directly comparable

  • OCRBench V2

    Qwen3.5 397B—
    Qwen3.8 Max74.2%
    Source

    Not directly comparable

  • CC-OCR

    Qwen3.5 397B—
    Qwen3.8 Max79.6%
    Source

    Not directly comparable

  • RealWorldQA

    Qwen3.5 397B—
    Qwen3.8 Max88.0%
    Source

    Not directly comparable

  • ERQA

    Qwen3.5 397B—
    Qwen3.8 Max77.8%
    Source

    Not directly comparable

  • SimpleVQA

    Qwen3.5 397B—
    Qwen3.8 Max75.0%
    Source

    Not directly comparable

  • PerceptionBench

    Qwen3.5 397B—
    Qwen3.8 Max63.5%
    Source

    Not directly comparable

  • Video-MME (with subtitle)

    Qwen3.5 397B—
    Qwen3.8 Max90.4%
    Source

    Not directly comparable

  • MMVU

    Qwen3.5 397B—
    Qwen3.8 Max82.4%
    Source

    Not directly comparable

  • MLVU (M-Avg)

    Qwen3.5 397B—
    Qwen3.8 Max90.8%
    Source

    Not directly comparable

  • LVBench

    Qwen3.5 397B—
    Qwen3.8 Max81.8%
    Source

    Not directly comparable

Instruction following

  • IFEval

    Qwen3.5 397B92.6%
    Source
    Qwen3.8 Max—

    Not directly comparable

  • IFBench

    Qwen3.5 397B—
    Qwen3.8 Max82.8%
    Source

    Not directly comparable

Questions

Which is better, Qwen3.5 397B or Qwen3.8 Max?

Qwen3.8 Max has the higher public score, 71.76 versus 54.87, and the 90% score intervals do not overlap. That is the clearest overall quality signal in the current public data.

Which is better for coding, Qwen3.5 397B or Qwen3.8 Max?

Qwen3.8 Max scores higher for coding on the public lane, 60.8 to 50.4. Qwen3.5 397B is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which is better for agentic tasks, Qwen3.5 397B or Qwen3.8 Max?

Qwen3.8 Max scores higher for agentic tasks on the public lane, 67.3 to 46.8. Qwen3.5 397B is scored on Estimated evidence for agentic tasks, so the reading is directional rather than like-for-like. BenchLM does not name a winner for agentic tasks from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which costs less, Qwen3.5 397B or Qwen3.8 Max?

Both models do not have comparable published API token rates, so this page does not name a universal price winner.

Which has the larger context window, Qwen3.5 397B or Qwen3.8 Max?

Qwen3.8 Max has the larger documented context window: 1M, compared with 128K.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated September 15, 2026

Watch Qwen3.5 397B vs Qwen3.8 Max

One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.

Read a sample issue

Join 2,000+ readers.