Skip to main content
BenchLM
Advertise

Model comparison

Qwen3.8 Max vs Trinity-Large-Thinking

Updated August 3, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.

Qwen3.8 Max

Alibaba

65.4/100

Estimated · Public rank #31

90% interval 55.5–75.3

Trinity-Large-Thinking

Arcee AI

47.5/100

Supported · Public rank #134

90% interval 30.7–64.2

Qwen3.8 Max has the higher public score estimate, 65.4 versus 47.45, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

1 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.

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

    No shared weighted benchmark basis supports a winner.

    Confidence: limited

  • Agentic work

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

    Not enough matched evidence

    No shared weighted benchmark basis supports a winner.

    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

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

    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
1
Qwen3.8 Max only
51
Trinity-Large-Thinking only
4
Like-for-like categories
0 / 8

Category results, on a stated basis

Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.

Agentic

Not comparable
Qwen3.8 Max
86.1
Trinity-Large-Thinking
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

Coding

Not comparable
Qwen3.8 Max
67.7
Trinity-Large-Thinking
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

Reasoning

Not comparable
Qwen3.8 Max
78.3
Trinity-Large-Thinking
Not measured
Weighted basis
2 vs 0 rows
Reading
Not comparable

Knowledge

Not comparable
Qwen3.8 Max
50.2
Trinity-Large-Thinking
Not measured
Weighted basis
2 vs 0 rows
Reading
Not comparable

Math

Not comparable
Qwen3.8 Max
Not measured
Trinity-Large-Thinking
Not measured
Weighted basis
0 vs 0 rows
Reading
Not comparable

Multilingual

Not comparable
Qwen3.8 Max
Not measured
Trinity-Large-Thinking
Not measured
Weighted basis
0 vs 0 rows
Reading
Not comparable

Multimodal

Not comparable
Qwen3.8 Max
86.3
Trinity-Large-Thinking
Not measured
Weighted basis
2 vs 0 rows
Reading
Not comparable

Instruction following

Not comparable
Qwen3.8 Max
82.8
Trinity-Large-Thinking
Not measured
Weighted basis
1 vs 0 rows
Reading
Not comparable

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.

A shared-evidence shape is not available.

BenchLM does not draw a radar or infer missing axes when the matched evidence is too sparse.

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

Qwen3.8 Max
API rate not published
Fits in one request
Trinity-Large-Thinking
$0.0007
Fits in one request

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

Repository review

50K fresh input + 3K output tokens

Qwen3.8 Max
API rate not published
Fits in one request
Trinity-Large-Thinking
$0.0152
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

Qwen3.8 Max
API rate not published
Fits in one request
Cached-input rate unavailable
Trinity-Large-Thinking
$0.064
Fits in one request
Cached input priced at the published list-input rate

Trinity-Large-Thinking 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.

Qwen3.8 Max

No comparable hosted API rate

Alibaba Cloud Model Studio pricing

Trinity-Large-Thinking

Not published

Documented inputs

Qwen3.8 Max

Not sourced

Trinity-Large-Thinking

Not sourced

Documented outputs

Qwen3.8 Max

Not sourced

Trinity-Large-Thinking

Not sourced

Provider availability

Qwen3.8 Max

Not sourced

Trinity-Large-Thinking

Not sourced

Reasoning profile

Qwen3.8 Max

Reasoning

Trinity-Large-Thinking

Reasoning

Weight access

Qwen3.8 Max

Proprietary

Trinity-Large-Thinking

Open Weight

License

Qwen3.8 Max

Proprietary

Trinity-Large-Thinking

Open Weight

Release date

Qwen3.8 Max

2026-08-03

Trinity-Large-Thinking

2026-03-10

If you already use one of these models
Deployment change
The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
Quality signal
Qwen3.8 Max has the higher public score estimate, 65.4 versus 47.45, but the 90% score intervals 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 evidence56 rows

Agentic

  • Terminal-Bench 2.1

    Qwen3.8 Max86.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • CoWorkBench

    Qwen3.8 Max74.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • JobBench

    Qwen3.8 Max53.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • skillsBench

    Qwen3.8 Max70.2%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • Agents' Last Exam

    Qwen3.8 Max52.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • AutomationBench

    Qwen3.8 Max27.3%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • Toolathlon-Verified

    Qwen3.8 Max72.5%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • WideResearch

    Qwen3.8 Max81.9%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • HLE w/ tools

    Qwen3.8 Max56.2%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • OSWorld-Verified

    Qwen3.8 Max86.1%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • OSWorld 2.0

    Qwen3.8 Max19.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • WebArena-Verified

    Qwen3.8 Max66.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • AndroidWorld

    Qwen3.8 Max85.3%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MobileWorld

    Qwen3.8 Max77.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • Gert Labs

    Qwen3.8 Max—
    Trinity-Large-Thinking32.55%
    Source

    Not directly comparable

Coding

  • Terminal-Bench 2.1

    Qwen3.8 Max86.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • SWE-bench Pro

    Qwen3.8 Max67.7%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • deepSwe

    Qwen3.8 Max56.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • NL2Repo

    Qwen3.8 Max55.9%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • FrontierSWE

    Qwen3.8 Max73.5%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MLS-Bench Lite

    Qwen3.8 Max41.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • PaperBench

    Qwen3.8 Max93.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • SWE-bench Verified*

    Qwen3.8 Max—
    Trinity-Large-Thinking63.2%
    Source

    Not directly comparable

Reasoning

  • MRCRv2

    Qwen3.8 Max92.9%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • LongBench v2

    Qwen3.8 Max66.3%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

Knowledge

  • GPQA

    Qwen3.8 Max92.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • GPQA-D

    Qwen3.8 Max92.6%
    Source
    Trinity-Large-Thinking76.3%
    Source

    Qwen3.8 Max leads this result

  • HLE

    Qwen3.8 Max43.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • HLE w/o tools

    Qwen3.8 Max43.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MMLU-Pro (Arcee)

    Qwen3.8 Max—
    Trinity-Large-Thinking83.4%
    Source

    Not directly comparable

Math

  • AIME25 (Arcee)

    Qwen3.8 Max—
    Trinity-Large-Thinking96.3%
    Source

    Not directly comparable

Multimodal

  • MMMU-Pro

    Qwen3.8 Max82.3%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MathVision

    Qwen3.8 Max95.2%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MathVision w/ Python

    Qwen3.8 Max97.7%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • BabyVision

    Qwen3.8 Max82.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • BabyVision w/ Python

    Qwen3.8 Max91.3%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • ZeroBench

    Qwen3.8 Max24.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • ZeroBench w/ Python

    Qwen3.8 Max49.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MedXpertQA (MM)

    Qwen3.8 Max80.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • ScreenSpot Pro

    Qwen3.8 Max84.5%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • Vision2Web

    Qwen3.8 Max69.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • CharXiv w/o tools

    Qwen3.8 Max88.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • CharXiv

    Qwen3.8 Max93.5%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • OmniDocBench 1.5

    Qwen3.8 Max92.1%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • OCRBench V2

    Qwen3.8 Max74.2%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • CC-OCR

    Qwen3.8 Max79.6%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • RealWorldQA

    Qwen3.8 Max88.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • ERQA

    Qwen3.8 Max77.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • SimpleVQA

    Qwen3.8 Max75.0%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • PerceptionBench

    Qwen3.8 Max63.5%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • Video-MME (with subtitle)

    Qwen3.8 Max90.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • VideoMMMU

    Qwen3.8 Max88.7%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MMVU

    Qwen3.8 Max82.4%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • MLVU (M-Avg)

    Qwen3.8 Max90.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

  • LVBench

    Qwen3.8 Max81.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

Instruction following

  • IFBench

    Qwen3.8 Max82.8%
    Source
    Trinity-Large-Thinking—

    Not directly comparable

Frequently asked questions

Which is better, Qwen3.8 Max or Trinity-Large-Thinking?

Qwen3.8 Max has the higher public score estimate, 65.4 versus 47.45, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.

Which is better for coding, Qwen3.8 Max or Trinity-Large-Thinking?

The published evidence does not provide a shared weighted coding basis for both models, so BenchLM does not name a coding winner.

Which is better for agentic tasks, Qwen3.8 Max or Trinity-Large-Thinking?

The published evidence does not provide a shared weighted agentic tasks basis for both models, so BenchLM does not name a agentic tasks winner.

Which costs less, Qwen3.8 Max or Trinity-Large-Thinking?

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, Qwen3.8 Max or Trinity-Large-Thinking?

Qwen3.8 Max has the larger documented context window: 1M, compared with 512K.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated August 3, 2026

Watch Qwen3.8 Max vs Trinity-Large-Thinking

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

Read a sample issue

Join 2,000+ readers.