ZMIME
Comparison · 3 models · Updated Nov 22, 2026

GLM-5.2 vs Qwen3.8 Max 0902 vs Muse Spark 1.1

Too close to call on our weighted score (Muse Spark 1.1 66, GLM-5.2 65, Qwen3.8 Max 0902 64). The right pick depends on what you value most.Too close to call on our weighted score (Qwen3.8 Max 0902 80, Muse Spark 1.1 80, GLM-5.2 78). The right pick depends on what you value most.

  1. Z.ai (Zhipu)

    GLM-5.2

    Released Jun 13, 2026

    65/100
    • ECI153.1
    • Price$1.40 / $4.40
    • Context1M
  2. Alibaba (Qwen)

    Qwen3.8 Max 0902

    Released Sep 2, 2026

    64/100
    • ECI156.0
    • Price$2.00 / $6.00
    • Context1M
  3. Meta

    Muse Spark 1.1

    Released Jul 9, 2026

    66/100
    • ECI154.9
    • Price$1.25 / $4.25
    • Context1.05M
  4. Z.ai (Zhipu)

    GLM-5.2

    Released Jun 13, 2026

    78/100
    • ECI153.1
    • Price$1.40 / $4.40
    • Context1M
  5. Alibaba (Qwen)

    Qwen3.8 Max 0902

    Released Sep 2, 2026

    80/100
    • ECI156.0
    • Price$2.00 / $6.00
    • Context1M
  6. Meta

    Muse Spark 1.1

    Released Jul 9, 2026

    80/100
    • ECI154.9
    • Price$1.25 / $4.25
    • Context1.05M
01 — Verdict

Too close to call

It is close. Our weighted score puts them within 2 points (Muse Spark 1.1 66/100, GLM-5.2 65/100, Qwen3.8 Max 0902 64/100), so choose by what matters most for your work: Qwen3.8 Max 0902 for raw capability, Muse Spark 1.1 on price and Muse Spark 1.1 for long inputs. The score weighs capability 59%, price 29%, context window 12%.It is close. Our weighted score puts them within a point (Qwen3.8 Max 0902 80/100, Muse Spark 1.1 80/100, GLM-5.2 78/100), so choose by what matters most for your work: Qwen3.8 Max 0902 for raw capability, Muse Spark 1.1 on price and Muse Spark 1.1 for long inputs. The score weighs capability 83%, context window 17%.

  • CapabilityQwen3.8 Max 0902Capabilities Index (ECI): Qwen3.8 Max 0902 156.0 · Muse Spark 1.1 154.9 · GLM-5.2 153.1
  • Lowest priceMuse Spark 1.1Muse Spark 1.1 $2.00 · GLM-5.2 $2.15 · Qwen3.8 Max 0902 $3.00 per 1M tokens (3:1 blend)
  • Longest contextMuse Spark 1.1Muse Spark 1.1 1,048,576 · GLM-5.2 1,000,000 · Qwen3.8 Max 0902 1,000,000 tokens
  • Widest inputsQwen3.8 Max 0902 and Muse Spark 1.1GLM-5.2: Text · Qwen3.8 Max 0902: Text, Images, PDFs, Video · Muse Spark 1.1: Text, Images, PDFs, Video
  • Self-hostingGLM-5.2Publishes downloadable weights
How the score is built
MeasureWeightGLM-5.2Qwen3.8 Max 0902Muse Spark 1.1
CapabilityCapabilities Index (ECI)59%818483
Price29%332634
Context window12%646465
Overall100%65/10064/10066/100

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

How the score is built
MeasureWeightGLM-5.2Qwen3.8 Max 0902Muse Spark 1.1
CapabilityCapabilities Index (ECI)83%818483
Context window17%646465
Overall100%78/10080/10080/100

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.

GLM-5.2 vs Qwen3.8 Max 0902 vs Muse Spark 1.1 specifications side by side
SpecificationGLM-5.2Z.ai (Zhipu)Qwen3.8 Max 0902Alibaba (Qwen)Muse Spark 1.1Meta
Capability
Capabilities Index (ECI)153.1156.0 (best)154.9
ECI rank#55 of 159#42 of 159 (best)#49 of 159
ECI confidence interval151.3–154.9153.9–158.3152.4–157.7
ECI evaluation dateJun 16, 2026Sep 1, 2026Jul 9, 2026
Documented capabilitiesOpen flagship GLM for long-horizon coding agents and million-token context work2026-09-02 upgraded snapshot of Qwen3.8 Max with stronger coding, collaborative agents, and multimodal document understandingMuse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.
Model parametersNot publishedNot publishedNot published
Output modalitiesTextTextText
GPQA DiamondGraduate-level science questions91.9%max92.3%xhigh (best)Not published
FrontierMath Tiers 1–3Research-level mathematics59.2%max65.6%xhigh (best)Not published
OTIS Mock AIME 2024–2025Competition mathematics86.4%max100%xhigh (best)Not published
SWE-bench VerifiedFixing real GitHub issues78.7%maxNot publishedNot published
SimpleQA VerifiedShort factual questions34.2%max47.3%xhigh57.8% (best)
Publisher-reported benchmarks
SWE-Bench ProPublisher-reported result62.1resolve rateReported resultJun 16, 2026Not published61.5resolve rateReported resultJul 9, 2026
Terminal-Bench 2.1Publisher-reported result82.7success rateReported resultJun 16, 2026Not published80success rateReported resultJul 9, 2026
FrontierSWEPublisher-reported result74.4dominanceReported resultJun 16, 2026Not publishedNot published
Humanity's Last Exam · text-only subsetPublisher-reported result40.5accuracyReported resultJun 16, 2026Not publishedNot published
Humanity's Last Exam · text-only subsetwith tools54.7accuracyReported resultJun 16, 2026Not publishedNot published
CritPtPublisher-reported result20.9accuracyReported resultJun 16, 2026Not publishedNot published
AIME 2026Publisher-reported result99.2accuracyReported resultJun 16, 2026Not publishedNot published
HMMT November 2025Publisher-reported result94.4accuracyReported resultJun 16, 2026Not publishedNot published
HMMT February 2026Publisher-reported result92.5accuracyReported resultJun 16, 2026Not publishedNot published
IMOAnswerBenchPublisher-reported result91accuracyReported resultJun 16, 2026Not publishedNot published
GPQA DiamondPublisher-reported result91.2accuracyReported resultJun 16, 2026Not publishedNot published
NL2RepoPublisher-reported result48.9resolve rateReported resultJun 16, 2026Not publishedNot published
DeepSWEPublisher-reported result46.2resolve rateReported resultJun 16, 2026Not publishedNot published
Program BenchPublisher-reported result63.7scoreReported resultJun 16, 2026Not publishedNot published
PostTrainBenchPublisher-reported result34.3scoreReported resultJun 16, 2026Not publishedNot published
SWE MarathonPublisher-reported result13resolve rateReported resultJun 16, 2026Not publishedNot published
MCP Atlas · public subsetPublisher-reported result76.8scoreReported resultJun 16, 2026Not publishedNot published
Tool-DecathlonPublisher-reported result48.2scoreReported resultJun 16, 2026Not publishedNot published
DeepSWE 1.1Publisher-reported resultNot publishedNot published53.3resolve rateReported resultJul 9, 2026
MCP AtlasPublisher-reported resultNot publishedNot published88.1success rateReported resultJul 9, 2026
JobBenchPublisher-reported resultNot publishedNot published54.7success rateReported resultJul 9, 2026
Toolathlon-VerifiedPublisher-reported resultNot publishedNot published75.6success rateReported resultJul 9, 2026
Humanity's Last Examwith toolsNot publishedNot published62.1accuracyReported resultJul 9, 2026
OSWorld-VerifiedPublisher-reported resultNot publishedNot published80.8success rateReported resultJul 9, 2026
Finance Agent v2Publisher-reported resultNot publishedNot published57.2accuracyReported resultJul 9, 2026
CharXiv ReasoningPublisher-reported resultNot publishedNot published88.4accuracyReported resultJul 9, 2026
BabyVisionPublisher-reported resultNot publishedNot published76.3accuracyReported resultJul 9, 2026
Price per million tokens
Input$1.40$2.00$1.25 (best)
Output$4.40$6.00$4.25 (best)
Cached input$0.26Not published$0.15 (best)
Blended (3:1)$2.15$3.00$2.00 (best)
Long-context rateSame rateSame rateSame rate
Price sourceOfficial Z.AI APIMedian of 9 providersOfficial Meta API
Limits
Context window1,000,000 tokens1,000,000 tokens1,048,576 tokens (best)
Max inputNot publishedNot publishedNot published
Max output131,072 tokens131,072 tokens131,072 tokens
Inputs and features
TextYesYesYes
ImagesNoYesYes
PDFsNoYesYes
AudioNoNoNo
VideoNoYesYes
ReasoningYeshigh · maxYesYesminimal · low · medium · high · xhigh
Tool callingYesYesYes
Structured outputYesNot publishedYes
File attachmentsNoYesYes
Temperature controlYesYesYes
Reasoning controleffortalwayseffort
Availability
WeightsOpenProprietaryProprietary
API model IDglm-5.2Not publishedmuse-spark-1.1
API providers88 (best)915
ReleasedJun 13, 2026Sep 2, 2026Jul 9, 2026
Knowledge cutoffNot publishedNot publishedNot published
Publisher documentationPublisher documentation · Model weightsNot publishedPublisher documentation
Source metadata updatedJun 13, 2026Sep 2, 2026Jul 9, 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.

  • GLM-5.2$22.80
  • Qwen3.8 Max 0902$32.00
  • Muse Spark 1.1$21.00
04 — Questions

Which should you choose?

Which is better: GLM-5.2, Qwen3.8 Max 0902 or Muse Spark 1.1?

It is close. Our weighted score puts them within 2 points (Muse Spark 1.1 66/100, GLM-5.2 65/100, Qwen3.8 Max 0902 64/100), so choose by what matters most for your work: Qwen3.8 Max 0902 for raw capability, Muse Spark 1.1 on price and Muse Spark 1.1 for long inputs. The score weighs capability 59%, price 29%, context window 12%.It is close. Our weighted score puts them within a point (Qwen3.8 Max 0902 80/100, Muse Spark 1.1 80/100, GLM-5.2 78/100), so choose by what matters most for your work: Qwen3.8 Max 0902 for raw capability, Muse Spark 1.1 on price and Muse Spark 1.1 for long inputs. The score weighs capability 83%, context window 17%.

Which is cheaper, GLM-5.2, Qwen3.8 Max 0902 or Muse Spark 1.1?

Muse Spark 1.1 is cheaper at $1.25 input / $4.25 output per million tokens (official Meta API price). GLM-5.2 costs $1.40 input / $4.40 output per million tokens (official Z.AI API price); Qwen3.8 Max 0902 costs $2.00 input / $6.00 output per million tokens (median across 9 API providers). At a typical mix of three input tokens to one output token, that is $2.00 per million tokens for Muse Spark 1.1 versus $2.15 for GLM-5.2 (1.1× as much) and $3.00 for Qwen3.8 Max 0902 (1.5× as much).

Which scores higher on benchmarks?

Qwen3.8 Max 0902 scores higher on the Capabilities Index (ECI): Qwen3.8 Max 0902 156.0 (#42 of 159), Muse Spark 1.1 154.9 (#49 of 159) and GLM-5.2 153.1 (#55 of 159). The confidence ranges of the top two overlap (153.9–158.3 vs 152.4–157.7), so treat the gap as small. On individual benchmarks: SimpleQA Verified — Muse Spark 1.1 57.8%, Qwen3.8 Max 0902 47.3%, GLM-5.2 34.2%.

Which is better for coding?

There are no published SWE-bench Verified results for Qwen3.8 Max 0902 and Muse Spark 1.1 yet, so there is no like-for-like coding score. On overall capability, Qwen3.8 Max 0902 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?

Muse Spark 1.1 has the largest context window at 1,048,576 tokens, against 1,000,000 for GLM-5.2 and 1,000,000 for Qwen3.8 Max 0902. Maximum output per response: GLM-5.2 up to 131,072, Qwen3.8 Max 0902 up to 131,072, Muse Spark 1.1 up to 131,072 tokens.

Which can read images, PDFs, audio or video?

GLM-5.2 accepts text; Qwen3.8 Max 0902 accepts text, images, PDFs and video; Muse Spark 1.1 accepts text, images, PDFs and video. Qwen3.8 Max 0902 handles the widest range of inputs.

Are any of these open source?

GLM-5.2 publishes its weights and can be self-hosted; Qwen3.8 Max 0902 and Muse Spark 1.1 is proprietary.

Which is newer?

Qwen3.8 Max 0902 is the newest, released Sep 2, 2026. Muse Spark 1.1 came out Jul 9, 2026; GLM-5.2 came out Jun 13, 2026.

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.