GPT-5.6 Luna vs Qwen3.8 Max

Decision reading

Qwen3.8 Max has the higher public score estimate, 71.76 versus 64.66, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

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

Model A

GPT-5.6 Luna

OpenAI

64.66/100

Estimated · Public rank #40

90% interval 58.9–70.4

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 X LinkedInSocial card CSV JSON

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.

Coding work
Code generation, repair, and software-engineering tasks
GPT-5.6 Luna
GPT-5.6 Luna leads on the public coding lane, 66.8 to 60.8, with Supported evidence for both models, although the 90% intervals overlap.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Qwen3.8 Max
Qwen3.8 Max leads on the public agentic lane, 67.3 to 56.6, with Supported evidence for both models, although the 90% intervals overlap.
Confidence: limited
Long documents
Prompts that approach the documented context limit
GPT-5.6 Luna
GPT-5.6 Luna has the larger documented context window.
Confidence: documented

Show secondary and unsupported calls

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: listed-rates
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.

19 GPT-5.6 Luna only

11 Shared

49 Qwen3.8 Max only

Shared results: 11
GPT-5.6 Luna only: 19
Qwen3.8 Max only: 49
Like-for-like categories: 3 / 8

1 category rests 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.

Agentic

Like-for-like

GPT-5.6 Luna: 56.6; Supported · #33/153
Qwen3.8 Max: 67.3; Supported · #9/153
Basis: BenchAlign lane · 9 vs 15 public rows
Reading: Qwen3.8 Max leads · intervals overlap

Coding

Like-for-like

GPT-5.6 Luna: 66.8; Supported · #9/152
Qwen3.8 Max: 60.8; Supported · #20/152
Basis: BenchAlign lane · 8 vs 12 public rows
Reading: GPT-5.6 Luna leads · intervals overlap

Knowledge

Like-for-like

GPT-5.6 Luna: 64.0; Supported · #27/183
Qwen3.8 Max: 68.8; Supported · #17/183
Basis: BenchAlign lane · 6 vs 6 public rows
Reading: Qwen3.8 Max leads · intervals overlap

Multimodal

Directional only

GPT-5.6 Luna: 67.1; #21/48
Qwen3.8 Max: 87.4; #5/48
Basis: Provisional lane · 1 vs 2 weighted rows
Reading: Directional only

Reasoning

Not comparable

GPT-5.6 Luna: 51.2; #19/20
Qwen3.8 Max: 86.5; Unranked · 2 rankable rows
Basis: Provisional lane · 2 vs 2 weighted rows
Reading: Not comparable

Math

Not comparable

GPT-5.6 Luna: 94.4; Unranked · 3 rankable rows
Qwen3.8 Max: Not ranked
Basis: Provisional lane · 2 vs 0 weighted rows
Reading: Not comparable

Multilingual

Not comparable

GPT-5.6 Luna: Not ranked
Qwen3.8 Max: Not ranked
Basis: Provisional lane · 0 vs 0 weighted rows
Reading: Not comparable

Instruction following

Not comparable

GPT-5.6 Luna: Not ranked
Qwen3.8 Max: 90.7; #18/123
Basis: Provisional lane · 0 vs 1 weighted rows
Reading: Not comparable

Category scores with the server-provided evidence basis for GPT-5.6 Luna and Qwen3.8 Max
Category	GPT-5.6 Luna	Qwen3.8 Max	Basis	Reading
Agentic	56.6Supported · #33/153	67.3Supported · #9/153	Like-for-likeBenchAlign lane · 9 vs 15 public rows	Qwen3.8 Max leads · intervals overlap
Coding	66.8Supported · #9/152	60.8Supported · #20/152	Like-for-likeBenchAlign lane · 8 vs 12 public rows	GPT-5.6 Luna leads · intervals overlap
Knowledge	64.0Supported · #27/183	68.8Supported · #17/183	Like-for-likeBenchAlign lane · 6 vs 6 public rows	Qwen3.8 Max leads · intervals overlap
Multimodal	67.1#21/48	87.4#5/48	Directional onlyProvisional lane · 1 vs 2 weighted rows	Directional only
Reasoning	51.2#19/20	86.5Unranked · 2 rankable rows	Not comparableProvisional lane · 2 vs 2 weighted rows	Not comparable
Math	94.4Unranked · 3 rankable rows	Not ranked	Not comparableProvisional lane · 2 vs 0 weighted rows	Not comparable
Multilingual	Not ranked	Not ranked	Not comparableProvisional lane · 0 vs 0 weighted rows	Not comparable
Instruction following	Not ranked	90.7#18/123	Not comparableProvisional lane · 0 vs 1 weighted rows	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.

OSWorld 2.0
Agentic
GPT-5.6 Luna: 45.6%Qwen3.8 Max: 19.4%Normalized gap 26.2GPT-5.6 Luna source Qwen3.8 Max source
SWE-bench Pro
Coding
GPT-5.6 Luna: 62.7%Qwen3.8 Max: 67.7%Normalized gap 5.0GPT-5.6 Luna source Qwen3.8 Max source
MMMU-Pro
Multimodal
GPT-5.6 Luna: 78.4%Qwen3.8 Max: 82.3%Normalized gap 3.9GPT-5.6 Luna source Qwen3.8 Max source
MMLU-Pro (Vals)
Knowledge
GPT-5.6 Luna: 86.0%Qwen3.8 Max: 88.6%Normalized gap 2.6GPT-5.6 Luna source Qwen3.8 Max source
GPQA
Knowledge
GPT-5.6 Luna: 92.3%Qwen3.8 Max: 92.6%Normalized gap 0.3GPT-5.6 Luna source Qwen3.8 Max source

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

GPT-5.6 Luna: $0.0008; 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

GPT-5.6 Luna: $0.0136; 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

GPT-5.6 Luna: $0.02; Fits in one request
Qwen3.8 Max: API rate not published; Fits in one request; Cached-input rate unavailable

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.

SpecificationGPT-5.6 LunaQwen3.8 Max

Context window

Maximum documented context; output-token limits may be lower.

GPT-5.6 Luna

1.05M

OpenAI model catalog

Qwen3.8 Max

Alibaba Cloud Model Studio pricing

API model ID

GPT-5.6 Luna

gpt-5.6-luna

OpenAI model catalog

Qwen3.8 Max

qwen3.8-max

Alibaba Cloud Model Studio pricing

Cached-input rate

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

GPT-5.6 Luna

$0.02 per 1M cached input tokens

OpenAI pricing

Qwen3.8 Max

No comparable hosted API rate

Alibaba Cloud Model Studio pricing

Documented inputs

GPT-5.6 Luna

text, image

OpenAI model catalog

Qwen3.8 Max

Not sourced

Documented outputs

GPT-5.6 Luna

text

OpenAI model catalog

Qwen3.8 Max

Not sourced

Provider availability

GPT-5.6 Luna

Generally Available · OpenAI Responses API

OpenAI model catalog

Qwen3.8 Max

Not sourced

Reasoning profile

GPT-5.6 Luna

Reasoning

Qwen3.8 Max

Reasoning

Weight access

GPT-5.6 Luna

Proprietary

Qwen3.8 Max

Open Weight

License

GPT-5.6 Luna

Proprietary

Qwen3.8 Max

Open Weight

Release date

GPT-5.6 Luna

2026-07-09

Qwen3.8 Max

2026-08-03

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, 71.76 versus 64.66, but the 90% score intervals overlap.
Workload cost: A complete comparable API-rate estimate is not available for both models.
Context tradeoff: GPT-5.6 Luna has the larger documented window (1.05M).

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 evidence79 rows

Agentic

Terminal-Bench 3.0
GPT-5.6 Luna14.3%
Source
Qwen3.8 Max—
Not directly comparable
Terminal-Bench 2.0
GPT-5.6 Luna84.7%
Source
Qwen3.8 Max—
Not directly comparable
BrowseComp
GPT-5.6 Luna83.3%
Source
Qwen3.8 Max—
Not directly comparable
OSWorld 2.0
GPT-5.6 Luna45.6%
Source
Qwen3.8 Max19.4%
Source
GPT-5.6 Luna leads this result
CyberGym
GPT-5.6 Luna77.9%
Source
Qwen3.8 Max—
Not directly comparable
ExploitGym
GPT-5.6 Luna12.4%
Source
Qwen3.8 Max—
Not directly comparable
Toolathlon
GPT-5.6 Luna53.4%
Source
Qwen3.8 Max—
Not directly comparable
Terminal-Bench 2.1 (Vals)
GPT-5.6 Luna79.0%
Source
Qwen3.8 Max67.4%
Source
GPT-5.6 Luna leads this result
ApprenticeBench
GPT-5.6 Luna7%
Source
Qwen3.8 Max—
Not directly comparable
Terminal-Bench 2.1
GPT-5.6 Luna—
Qwen3.8 Max86.6%
Source
Not directly comparable
CoWorkBench
GPT-5.6 Luna—
Qwen3.8 Max74.8%
Source
Not directly comparable
JobBench
GPT-5.6 Luna—
Qwen3.8 Max53.4%
Source
Not directly comparable
skillsBench
GPT-5.6 Luna—
Qwen3.8 Max70.2%
Source
Not directly comparable
Agents' Last Exam
GPT-5.6 Luna—
Qwen3.8 Max52.4%
Source
Not directly comparable
AutomationBench
GPT-5.6 Luna—
Qwen3.8 Max27.3%
Source
Not directly comparable
Toolathlon-Verified
GPT-5.6 Luna—
Qwen3.8 Max72.5%
Source
Not directly comparable
WideResearch
GPT-5.6 Luna—
Qwen3.8 Max81.9%
Source
Not directly comparable
HLE w/ tools
GPT-5.6 Luna—
Qwen3.8 Max56.2%
Source
Not directly comparable
OSWorld-Verified
GPT-5.6 Luna—
Qwen3.8 Max86.1%
Source
Not directly comparable
WebArena-Verified
GPT-5.6 Luna—
Qwen3.8 Max66.8%
Source
Not directly comparable
AndroidWorld
GPT-5.6 Luna—
Qwen3.8 Max85.3%
Source
Not directly comparable
MobileWorld
GPT-5.6 Luna—
Qwen3.8 Max77.8%
Source
Not directly comparable

Coding

SWE-bench Pro
GPT-5.6 Luna62.7%
Source
Qwen3.8 Max67.7%
Source
Qwen3.8 Max leads this result
Terminal-Bench 2.0
GPT-5.6 Luna84.7%
Source
Qwen3.8 Max—
Not directly comparable
DeepSWE
GPT-5.6 Luna67.2%
Source
Qwen3.8 Max56.6%
Source
GPT-5.6 Luna leads this result
FrontierCode 1.1 Extended
GPT-5.6 Luna55.1%
Source
Qwen3.8 Max—
Not directly comparable
cursorBench32
GPT-5.6 Luna61.1%
Source
Qwen3.8 Max—
Not directly comparable
VulcanBench v3
Shared source
GPT-5.6 Luna85.5%
Qwen3.8 Max81.2%
GPT-5.6 Luna leads this result
SWE-bench (Vals)
GPT-5.6 Luna93.0%
Source
Qwen3.8 Max85.6%
Source
GPT-5.6 Luna leads this result
cursorBench40
GPT-5.6 Luna35.9%
Source
Qwen3.8 Max—
Not directly comparable
Terminal-Bench 2.1
GPT-5.6 Luna—
Qwen3.8 Max86.6%
Source
Not directly comparable
NL2Repo
GPT-5.6 Luna—
Qwen3.8 Max55.9%
Source
Not directly comparable
FrontierSWE
GPT-5.6 Luna—
Qwen3.8 Max73.5%
Source
Not directly comparable
MLS-Bench Lite
GPT-5.6 Luna—
Qwen3.8 Max41.0%
Source
Not directly comparable
PaperBench
GPT-5.6 Luna—
Qwen3.8 Max93.0%
Source
Not directly comparable
OpenHarmony Bench
GPT-5.6 Luna—
Qwen3.8 Max60.8%
Source
Not directly comparable
FrontierSWE v2
GPT-5.6 Luna—
Qwen3.8 Max15.8%
Source
Not directly comparable
LiveCodeBench (Vals)
GPT-5.6 Luna—
Qwen3.8 Max87.9%
Source
Not directly comparable

Reasoning

ARC-AGI-2
GPT-5.6 Luna59.5%
Source
Qwen3.8 Max—
Not directly comparable
ARC-AGI-3
GPT-5.6 Luna0.2%
Source
Qwen3.8 Max—
Not directly comparable
MRCRv2
GPT-5.6 Luna—
Qwen3.8 Max92.9%
Source
Not directly comparable
LongBench v2
GPT-5.6 Luna—
Qwen3.8 Max66.3%
Source
Not directly comparable

Knowledge

GPQA
GPT-5.6 Luna92.3%
Source
Qwen3.8 Max92.6%
Source
Qwen3.8 Max leads this result
GPQA-D
GPT-5.6 Luna92.3%
Source
Qwen3.8 Max92.6%
Source
Qwen3.8 Max leads this result
HealthBench Professional
GPT-5.6 Luna55.7%
Source
Qwen3.8 Max—
Not directly comparable
HealthBench Hard
GPT-5.6 Luna32.0%
Source
Qwen3.8 Max—
Not directly comparable
GPQA Diamond (Vals)
GPT-5.6 Luna91.7%
Source
Qwen3.8 Max93.7%
Source
Qwen3.8 Max leads this result
MMLU-Pro (Vals)
GPT-5.6 Luna86.0%
Source
Qwen3.8 Max88.6%
Source
Qwen3.8 Max leads this result
HLE
GPT-5.6 Luna—
Qwen3.8 Max43.6%
Source
Not directly comparable
HLE w/o tools
GPT-5.6 Luna—
Qwen3.8 Max43.6%
Source
Not directly comparable

Math

FrontierMath (legacy)
GPT-5.6 Luna78.6%
Source
Qwen3.8 Max—
Not directly comparable
FrontierMath v2 (Tiers 1-3)
GPT-5.6 Luna78.600%
Source
Qwen3.8 Max—
Not directly comparable
FrontierMath v2 (Tier 4)
GPT-5.6 Luna58.500%
Source
Qwen3.8 Max—
Not directly comparable

Multimodal

MMMU-Pro
GPT-5.6 Luna78.4%
Source
Qwen3.8 Max82.3%
Source
Qwen3.8 Max leads this result
MMMU-Pro w/ Python
GPT-5.6 Luna79.5%
Source
Qwen3.8 Max—
Not directly comparable
MathVision
GPT-5.6 Luna—
Qwen3.8 Max95.2%
Source
Not directly comparable
MathVision w/ Python
GPT-5.6 Luna—
Qwen3.8 Max97.7%
Source
Not directly comparable
BabyVision
GPT-5.6 Luna—
Qwen3.8 Max82.0%
Source
Not directly comparable
BabyVision w/ Python
GPT-5.6 Luna—
Qwen3.8 Max91.3%
Source
Not directly comparable
ZeroBench
GPT-5.6 Luna—
Qwen3.8 Max24.0%
Source
Not directly comparable
ZeroBench w/ Python
GPT-5.6 Luna—
Qwen3.8 Max49.0%
Source
Not directly comparable
MedXpertQA (MM)
GPT-5.6 Luna—
Qwen3.8 Max80.4%
Source
Not directly comparable
ScreenSpot Pro
GPT-5.6 Luna—
Qwen3.8 Max84.5%
Source
Not directly comparable
Vision2Web
GPT-5.6 Luna—
Qwen3.8 Max69.0%
Source
Not directly comparable
CharXiv w/o tools
GPT-5.6 Luna—
Qwen3.8 Max88.4%
Source
Not directly comparable
CharXiv
GPT-5.6 Luna—
Qwen3.8 Max93.5%
Source
Not directly comparable
OmniDocBench 1.5
GPT-5.6 Luna—
Qwen3.8 Max92.1%
Source
Not directly comparable
OCRBench V2
GPT-5.6 Luna—
Qwen3.8 Max74.2%
Source
Not directly comparable
CC-OCR
GPT-5.6 Luna—
Qwen3.8 Max79.6%
Source
Not directly comparable
RealWorldQA
GPT-5.6 Luna—
Qwen3.8 Max88.0%
Source
Not directly comparable
ERQA
GPT-5.6 Luna—
Qwen3.8 Max77.8%
Source
Not directly comparable
SimpleVQA
GPT-5.6 Luna—
Qwen3.8 Max75.0%
Source
Not directly comparable
PerceptionBench
GPT-5.6 Luna—
Qwen3.8 Max63.5%
Source
Not directly comparable
Video-MME (with subtitle)
GPT-5.6 Luna—
Qwen3.8 Max90.4%
Source
Not directly comparable
VideoMMMU
GPT-5.6 Luna—
Qwen3.8 Max88.7%
Source
Not directly comparable
MMVU
GPT-5.6 Luna—
Qwen3.8 Max82.4%
Source
Not directly comparable
MLVU (M-Avg)
GPT-5.6 Luna—
Qwen3.8 Max90.8%
Source
Not directly comparable
LVBench
GPT-5.6 Luna—
Qwen3.8 Max81.8%
Source
Not directly comparable

Instruction following

IFBench
GPT-5.6 Luna—
Qwen3.8 Max82.8%
Source
Not directly comparable

Questions

Which is better, GPT-5.6 Luna or Qwen3.8 Max?

Qwen3.8 Max has the higher public score estimate, 71.76 versus 64.66, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.

Which is better for coding, GPT-5.6 Luna or Qwen3.8 Max?

GPT-5.6 Luna leads the public coding lane, 66.8 to 60.8, with Supported evidence for both models, although the 90% intervals overlap.

Which is better for agentic tasks, GPT-5.6 Luna or Qwen3.8 Max?

Qwen3.8 Max leads the public agentic tasks lane, 67.3 to 56.6, with Supported evidence for both models, although the 90% intervals overlap.

Which costs less, GPT-5.6 Luna 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, GPT-5.6 Luna or Qwen3.8 Max?

GPT-5.6 Luna has the larger documented context window: 1.05M, compared with 1M.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated September 15, 2026

Watch GPT-5.6 Luna 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.

GPT-5.6 Luna vs Qwen3.8 Max

Which one for your work

Coding work

Agentic work

Long documents

Chat turn cost

Cache-heavy agent loop cost

Repository review cost

What is actually comparable

Category results, on a stated basis

Agentic

Coding

Knowledge

Multimodal

Reasoning

Math

Multilingual

Instruction following

Shape of the matched evidence

What each workload costs

Chat turn

Repository review

Cache-heavy agent loop

Specification differences

Context window

API model ID

Cached-input rate

Documented inputs

Documented outputs

Provider availability

Reasoning profile

Weight access

License

Release date

Benchmark evidence

Agentic

Coding

Reasoning

Knowledge

Math

Multimodal

Instruction following

Questions

Which is better, GPT-5.6 Luna or Qwen3.8 Max?

Which is better for coding, GPT-5.6 Luna or Qwen3.8 Max?

Which is better for agentic tasks, GPT-5.6 Luna or Qwen3.8 Max?

Which costs less, GPT-5.6 Luna or Qwen3.8 Max?

Which has the larger context window, GPT-5.6 Luna or Qwen3.8 Max?

Related comparisons

Watch GPT-5.6 Luna vs Qwen3.8 Max