ZMIME
Comparison · 3 models · Updated Nov 22, 2026

GPT-5.6 Luna vs GLM-5.3 vs DeepSeek V4.1 Flash

Too close to call on our weighted score (DeepSeek V4.1 Flash 76, GPT-5.6 Luna 76, GLM-5.3 67). The right pick depends on what you value most.Too close to call on our weighted score (GLM-5.3 81, DeepSeek V4.1 Flash 81, GPT-5.6 Luna 81). The right pick depends on what you value most.

  1. OpenAI

    GPT-5.6 Luna

    Released Jul 9, 2026

    76/100
    • ECI156.7
    • Price$0.20 / $1.20
    • Context1.05M
  2. Z.ai (Zhipu)

    GLM-5.3

    Released Aug 14, 2026Latest

    67/100
    • ECI157.1
    • Price$1.40 / $4.40
    • Context1M
  3. DeepSeek

    DeepSeek V4.1 Flash

    Released Sep 10, 2026

    76/100
    • ECI157.0
    • Price$0.222 / $1.04
    • Context1M
  4. OpenAI

    GPT-5.6 Luna

    Released Jul 9, 2026

    81/100
    • ECI156.7
    • Price$0.20 / $1.20
    • Context1.05M
  5. Z.ai (Zhipu)

    GLM-5.3

    Released Aug 14, 2026Latest

    81/100
    • ECI157.1
    • Price$1.40 / $4.40
    • Context1M
  6. DeepSeek

    DeepSeek V4.1 Flash

    Released Sep 10, 2026

    81/100
    • ECI157.0
    • Price$0.222 / $1.04
    • Context1M
01 — Verdict

Too close to call

It is close. Our weighted score puts them within a point (DeepSeek V4.1 Flash 76/100, GPT-5.6 Luna 76/100, GLM-5.3 67/100), so choose by what matters most for your work: GLM-5.3 for raw capability, DeepSeek V4.1 Flash on price and GPT-5.6 Luna for long inputs. The score weighs capability 59%, price 29%, context window 12%.It is close. Our weighted score puts them within a point (GLM-5.3 81/100, DeepSeek V4.1 Flash 81/100, GPT-5.6 Luna 81/100), so choose by what matters most for your work: GLM-5.3 for raw capability, DeepSeek V4.1 Flash on price and GPT-5.6 Luna for long inputs. The score weighs capability 83%, context window 17%.

  • CapabilityGLM-5.3Capabilities Index (ECI): GLM-5.3 157.1 · DeepSeek V4.1 Flash 157.0 · GPT-5.6 Luna 156.7
  • Lowest priceDeepSeek V4.1 FlashDeepSeek V4.1 Flash $0.426 · GPT-5.6 Luna $0.45 · GLM-5.3 $2.15 per 1M tokens (3:1 blend)
  • Longest contextGPT-5.6 LunaGPT-5.6 Luna 1,050,000 · GLM-5.3 1,000,000 · DeepSeek V4.1 Flash 1,000,000 tokens
  • Widest inputsGPT-5.6 LunaGPT-5.6 Luna: Text, Images, PDFs · GLM-5.3: Text · DeepSeek V4.1 Flash: Text, Images
  • Self-hostingGLM-5.3 and DeepSeek V4.1 FlashPublishes downloadable weights (MIT)
How the score is built
MeasureWeightGPT-5.6 LunaGLM-5.3DeepSeek V4.1 Flash
CapabilityCapabilities Index (ECI)59%848584
Price29%633365
Context window12%656464
Overall100%76/10067/10076/100

Left out because at least one model lacks the data: inputs & features. The remaining weights were rescaled.

How the score is built
MeasureWeightGPT-5.6 LunaGLM-5.3DeepSeek V4.1 Flash
CapabilityCapabilities Index (ECI)83%848584
Context window17%656464
Overall100%81/10081/10081/100

Left out because at least one model lacks the data: inputs & features. The remaining weights were rescaled.

02 — Side by side

Every spec in one table

Highlighted cells lead their row. Unavailable specifications are marked “Not published”. Publisher results include their evaluation settings and source.

GPT-5.6 Luna vs GLM-5.3 vs DeepSeek V4.1 Flash specifications side by side
SpecificationGPT-5.6 LunaOpenAIGLM-5.3Z.ai (Zhipu)DeepSeek V4.1 FlashDeepSeek
Capability
Capabilities Index (ECI)156.7157.1 (best)157.0
ECI rank#37 of 159#34 of 159 (best)#35 of 159
ECI confidence interval154.9–158.8155.0–159.4155.1–159.2
ECI evaluation dateJul 9, 2026Aug 14, 2026Sep 9, 2026
Documented capabilitiesCost-efficient GPT-5.6 model for fast, high-volume workloadsFlagship GLM model for long-horizon coding, agents, and complex project deliveryDeepSeek V4.1 Flash model for reasoning and agentic coding
Model parametersNot publishedNot publishedNot published
Output modalitiesTextTextText
GPQA DiamondGraduate-level science questions91.6%max (best)90.9%max89.8%max
FrontierMath Tiers 1–3Research-level mathematics82.1%max (best)68.8%max67.4%max
OTIS Mock AIME 2024–2025Competition mathematics98.3%max (best)91.1%max98.3%max (best)
SimpleQA VerifiedShort factual questions41.0%max41.0%max42.9%max (best)
Publisher-reported benchmarks
SWE-Bench ProPublisher-reported result62.7resolve rateReported resultJul 9, 2026Not publishedNot published
Terminal-Bench 2.1Publisher-reported result84.7success rateReported resultJul 9, 2026Not publishedNot published
DeepSWE 1.1Publisher-reported result67.2resolve rateReported resultJul 9, 2026Not publishedNot published
GPQA DiamondPublisher-reported result92.3accuracyReported resultJul 9, 2026Not publishedNot published
FrontierMath v2 · Tier 1-3Publisher-reported result78.6accuracyReported resultJul 9, 2026Not publishedNot published
BrowseCompPublisher-reported result83.3accuracyReported resultJul 9, 2026Not publishedNot published
OSWorld 2.0Publisher-reported result45.6success rateReported resultJul 9, 2026Not publishedNot published
MMMU Prono tools78.4accuracyReported resultJul 9, 2026Not publishedNot published
Agents' Last ExamPublisher-reported result50.3Reported resultJul 9, 2026Not publishedNot published
ToolathlonPublisher-reported result53.4success rateReported resultJul 9, 2026Not publishedNot published
Artificial Analysis Intelligence Index 4.1max51.2index scoreReported resultJul 9, 2026Not publishedNot published
Artificial Analysis Coding Agent Index 1.1max74.6index scoreReported resultJul 9, 2026Not publishedNot published
Terminal-Bench 2.1max effortNot published88.2pass@1Reported resultNot published
Terminal-Bench 3.0max effortNot published28.3avg@3Reported resultNot published
DeepSWE 1.1max effort; 6h timeout; 400K contextNot published66.9resolvedReported resultNot published
NL2Repomax effortNot published58scoreReported resultNot published
ProgramBenchmax effortNot published19almost solvedReported resultNot published
FrontierSWEmax effortNot published78.1dominance scoreReported resultAug 14, 2026Not published
SWE-Marathon 1.1max effort; modified anti-cheat checksNot published42.5scoreReported resultNot published
PostTrainBenchmax effort; 3 runs; modified external-API checksNot published39.8weighted averageReported resultNot published
CyberGymmax effortNot published84.5pass@1Reported resultNot published
ExploitGym · 869 tasksmax effort; 2h rescaled timeoutNot published105tasks solvedReported resultNot published
ExploitGym · 869 tasksmax effort; 6h rescaled timeoutNot published130tasks solvedReported resultNot published
ExploitBenchmax effort; union over 3 revisionsNot published54.4average coverageReported resultNot published
Toolathlon-Verifiedmax effortNot published73pass@1Reported resultNot published
AutomationBench 1.0.6max effortNot published48.2pass@1Reported resultNot published
Agents' Last Exammax effort; CLINot published28.5scoreReported resultNot published
Humanity's Last Exammax effort; with toolsNot published62.5accuracyReported resultNot published
GDPval-AA 2max effortNot published1769EloReported resultNot published
GPQA Diamondreasoning_effort=100Not publishedNot published90.9pass@1Reported result
MathArena Apexreasoning_effort=100Not publishedNot published65.6pass@1Reported result
Humanity's Last Exam · full setreasoning_effort=100Not publishedNot published36.8pass@1Reported result
Humanity's Last Exam · text-only subsetreasoning_effort=100Not publishedNot published39.1pass@1Reported result
Codeforcesreasoning_effort=100Not publishedNot published3471ratingReported result
Terminal-Bench 2.1reasoning_effort=100Not publishedNot published90.6pass@1Reported result
Terminal-Bench 3.0reasoning_effort=100Not publishedNot published30pass@1Reported result
Terminal-Bench 4.0reasoning_effort=100Not publishedNot published31.2pass@1Reported result
DeepSWE 1.1reasoning_effort=100Not publishedNot published74.2resolvedReported result
ProgramBenchreasoning_effort=100Not publishedNot published20.3almost@1Reported result
NL2Reporeasoning_effort=100Not publishedNot published64scoreReported result
CyberGymreasoning_effort=100Not publishedNot published88.1pass@1Reported result
ExploitGymreasoning_effort=100Not publishedNot published15.3pass@1Reported result
AutomationBenchreasoning_effort=100Not publishedNot published54.8pass@1Reported result
Agents' Last Examreasoning_effort=100Not publishedNot published31.8pass@1Reported result
SEC-Bench Proreasoning_effort=100Not publishedNot published62.8pass@1Reported result
Humanity's Last Examreasoning_effort=100; with toolsNot publishedNot published63.9pass@1Reported result
Chartographyreasoning_effort=100; with toolsNot publishedNot published78.9pass@1Reported result
BabyVisionreasoning_effort=100; with toolsNot publishedNot published89.6pass@1Reported result
ZeroBench · mainreasoning_effort=100; with toolsNot publishedNot published49pass@5Reported result
Price per million tokens
Input$0.20 (best)$1.40$0.222
Output$1.20$4.40$1.04 (best)
Cached input$0.02 (best)$0.26Not published
Blended (3:1)$0.45$2.15$0.426 (best)
Long-context rateOver 272K: $0.40 / $1.80Same rateSame rate
Price sourceOfficial OpenAI APIOfficial Z.AI APIMedian of 82 providers
Limits
Context window1,050,000 tokens (best)1,000,000 tokens1,000,000 tokens
Max input922,000 tokensNot publishedNot published
Max output128,000 tokens131,072 tokens384,000 tokens (best)
Inputs and features
TextYesYesYes
ImagesYesNoYes
PDFsYesNoNo
AudioNoNoNo
VideoNoNoNo
ReasoningYeslow · medium · high · xhigh · maxYeslow · high · maxYes
Tool callingYesYesYes
Structured outputYesYesNot published
File attachmentsYesNoYes
Temperature controlYesYesYes
Reasoning controlefforteffortalways
Availability
WeightsProprietaryOpenOpenMIT
API model IDgpt-5.6-lunaglm-5.3Not published
API providers57123 (best)86
ReleasedJul 9, 2026Aug 14, 2026Sep 10, 2026
Knowledge cutoffFeb 16, 2026Not publishedMay 2025
Publisher documentationPublisher documentationPublisher documentation · Model weightsModel weights
Source metadata updatedJul 9, 2026Aug 14, 2026Sep 10, 2026
Data sourcesmodels.dev · Epoch AImodels.dev · Epoch AImodels.dev · Epoch AI
Catalog checkedNov 22, 2026Nov 22, 2026Nov 22, 2026
03 — Cost

What would a month cost?

Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.

  • GPT-5.6 Luna$4.40
  • GLM-5.3$22.80
  • DeepSeek V4.1 Flash$4.29
04 — Questions

Which should you choose?

Which is better: GPT-5.6 Luna, GLM-5.3 or DeepSeek V4.1 Flash?

It is close. Our weighted score puts them within a point (DeepSeek V4.1 Flash 76/100, GPT-5.6 Luna 76/100, GLM-5.3 67/100), so choose by what matters most for your work: GLM-5.3 for raw capability, DeepSeek V4.1 Flash on price and GPT-5.6 Luna for long inputs. The score weighs capability 59%, price 29%, context window 12%.It is close. Our weighted score puts them within a point (GLM-5.3 81/100, DeepSeek V4.1 Flash 81/100, GPT-5.6 Luna 81/100), so choose by what matters most for your work: GLM-5.3 for raw capability, DeepSeek V4.1 Flash on price and GPT-5.6 Luna for long inputs. The score weighs capability 83%, context window 17%.

Which is cheaper, GPT-5.6 Luna, GLM-5.3 or DeepSeek V4.1 Flash?

DeepSeek V4.1 Flash is cheaper at $0.222 input / $1.04 output per million tokens (median across 82 API providers). GPT-5.6 Luna costs $0.20 input / $1.20 output per million tokens (official OpenAI API price); GLM-5.3 costs $1.40 input / $4.40 output per million tokens (official Z.AI API price). At a typical mix of three input tokens to one output token, that is $0.426 per million tokens for DeepSeek V4.1 Flash versus $0.45 for GPT-5.6 Luna (1.1× as much) and $2.15 for GLM-5.3 (5× as much).

Which scores higher on benchmarks?

GLM-5.3 scores higher on the Capabilities Index (ECI): GLM-5.3 157.1 (#34 of 159), DeepSeek V4.1 Flash 157.0 (#35 of 159) and GPT-5.6 Luna 156.7 (#37 of 159). The confidence ranges of the top two overlap (155.0–159.4 vs 155.1–159.2), so treat the gap as small. On individual benchmarks: GPQA Diamond — GPT-5.6 Luna 91.6%, GLM-5.3 90.9%, DeepSeek V4.1 Flash 89.8%; FrontierMath Tiers 1–3 — GPT-5.6 Luna 82.1%, GLM-5.3 68.8%, DeepSeek V4.1 Flash 67.4%; OTIS Mock AIME 2024–2025 — GPT-5.6 Luna 98.3%, DeepSeek V4.1 Flash 98.3%, GLM-5.3 91.1%; SimpleQA Verified — DeepSeek V4.1 Flash 42.9%, GPT-5.6 Luna 41.0%, GLM-5.3 41.0%.

Which is better for coding?

There are no published SWE-bench Verified results for GPT-5.6 Luna, GLM-5.3 and DeepSeek V4.1 Flash yet, so there is no like-for-like coding score. On overall capability, GLM-5.3 leads, which tends to carry over to coding, but test on your own codebase. All three support tool calling for agent workflows.

Which has the bigger context window?

GPT-5.6 Luna has the largest context window at 1,050,000 tokens, against 1,000,000 for GLM-5.3 and 1,000,000 for DeepSeek V4.1 Flash. Maximum output per response: GPT-5.6 Luna up to 128,000, GLM-5.3 up to 131,072, DeepSeek V4.1 Flash up to 384,000 tokens.

Which can read images, PDFs, audio or video?

GPT-5.6 Luna accepts text, images and PDFs; GLM-5.3 accepts text; DeepSeek V4.1 Flash accepts text and images. GPT-5.6 Luna handles the widest range of inputs.

Are any of these open source?

GLM-5.3 and DeepSeek V4.1 Flash publishes its weights (MIT) and can be self-hosted; GPT-5.6 Luna is proprietary.

Which is newer?

DeepSeek V4.1 Flash is the newest, released Sep 10, 2026. GLM-5.3 came out Aug 14, 2026; GPT-5.6 Luna came out Jul 9, 2026. Knowledge cutoff: GPT-5.6 Luna Feb 16, 2026, DeepSeek V4.1 Flash May 2025.

How do you decide the winner?

This rating uses capability (59%, independent benchmark results); price (29%, blended price per million tokens (3 input : 1 output), log scale); context window (12%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 22, 2026.This rating uses capability (83%, independent benchmark results); context window (17%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 22, 2026.