Z.aiLanguage model
GLM-4.6V-Flash
GLM-4.6V-Flash has no overall index value yet because there are too few independent benchmark results.
PopularityRank 182 of 785
45out of 100-3 vs. last month
based on search, usage and attention
Summary
GLM-4.6V-Flash brings the text, image, and video understanding of GLM-4.6V into an open 9B checkpoint. Its context is 128K tokens and its output remains text.
With up to 32,768 output tokens and API access, Flash is a distinct small model class rather than only a shortened-context configuration of the larger vision model. The weights are openly available.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 70 published benchmark results
Page 1 of 6
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
MedVision-V0 anatomy detection F1 (Show measurement, test conditions, and source)
- Source value
- 15.2
- Score
- 15.2
- Metric
- F1 score (%)
- Unit
- %
- Category
- medical-imaging-detection
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- MedVision-V0 public leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact MedVision-V0 leaderboard row GLM-4.6V-Flash (9B); corresponding public model repository https://huggingface.co/zai-org/GLM-4.6V-Flash. Scores are from the evaluator's current public Tables 2-4. The project's README says the leaderboard uses MedVision v1.0.0 annotations; it separately describes an open-weight limit of 1,000 samples per subtask and rounded totals in the public tables. No immutable evaluated weights revision, prompt, decoding setting, or reasoning mode is attached to the row.
- Context
- The source publishes rounded aggregate table denominators for detection and size, not an exact per-model scored N; no sampleCount is assigned for those cells. Distance and angle totals are explicit. No confidence interval is reported. The leaderboard documentation contains broader dataset-version statements, while its specific benchmark note says all leaderboard numbers use v1.0.0 annotations; the table does not bind an immutable model checkpoint or inference settings.
MedVision-V0 tumor/lesion detection F1 (Show measurement, test conditions, and source)
- Source value
- 7.5
- Score
- 7.5
- Metric
- F1 score (%)
- Unit
- %
- Category
- medical-imaging-detection
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- MedVision-V0 public leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact MedVision-V0 leaderboard row GLM-4.6V-Flash (9B); corresponding public model repository https://huggingface.co/zai-org/GLM-4.6V-Flash. Scores are from the evaluator's current public Tables 2-4. The project's README says the leaderboard uses MedVision v1.0.0 annotations; it separately describes an open-weight limit of 1,000 samples per subtask and rounded totals in the public tables. No immutable evaluated weights revision, prompt, decoding setting, or reasoning mode is attached to the row.
- Context
- The source publishes rounded aggregate table denominators for detection and size, not an exact per-model scored N; no sampleCount is assigned for those cells. Distance and angle totals are explicit. No confidence interval is reported. The leaderboard documentation contains broader dataset-version statements, while its specific benchmark note says all leaderboard numbers use v1.0.0 annotations; the table does not bind an immutable model checkpoint or inference settings.
MedVision-V0 tumor/lesion size-estimation success rate (Show measurement, test conditions, and source)
- Source value
- 95.2
- Score
- 95.2
- Metric
- success rate (%)
- Unit
- %
- Category
- medical-imaging-size-estimation
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- MedVision-V0 public leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact MedVision-V0 leaderboard row GLM-4.6V-Flash (9B); corresponding public model repository https://huggingface.co/zai-org/GLM-4.6V-Flash. Scores are from the evaluator's current public Tables 2-4. The project's README says the leaderboard uses MedVision v1.0.0 annotations; it separately describes an open-weight limit of 1,000 samples per subtask and rounded totals in the public tables. No immutable evaluated weights revision, prompt, decoding setting, or reasoning mode is attached to the row.
- Context
- The source publishes rounded aggregate table denominators for detection and size, not an exact per-model scored N; no sampleCount is assigned for those cells. Distance and angle totals are explicit. No confidence interval is reported. The leaderboard documentation contains broader dataset-version statements, while its specific benchmark note says all leaderboard numbers use v1.0.0 annotations; the table does not bind an immutable model checkpoint or inference settings.
MedVision-V0 distance-estimation success rate (Show measurement, test conditions, and source)
- Source value
- 99.7
- Score
- 99.7
- Metric
- success rate (%)
- Unit
- %
- Benchmark version
- Ceph-Bio-400 + FeTA24 distance task
- Category
- medical-imaging-distance-estimation
- Direction
- Higher is better
- Sample
- 1100
- Source type
- independent-evaluator
- Evaluator
- MedVision-V0 public leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact MedVision-V0 leaderboard row GLM-4.6V-Flash (9B); corresponding public model repository https://huggingface.co/zai-org/GLM-4.6V-Flash. Scores are from the evaluator's current public Tables 2-4. The project's README says the leaderboard uses MedVision v1.0.0 annotations; it separately describes an open-weight limit of 1,000 samples per subtask and rounded totals in the public tables. No immutable evaluated weights revision, prompt, decoding setting, or reasoning mode is attached to the row.
- Context
- The source publishes rounded aggregate table denominators for detection and size, not an exact per-model scored N; no sampleCount is assigned for those cells. Distance and angle totals are explicit. No confidence interval is reported. The leaderboard documentation contains broader dataset-version statements, while its specific benchmark note says all leaderboard numbers use v1.0.0 annotations; the table does not bind an immutable model checkpoint or inference settings.
MedVision-V0 angle-estimation success rate (Show measurement, test conditions, and source)
- Source value
- 99.1
- Score
- 99.1
- Metric
- success rate (%)
- Unit
- %
- Benchmark version
- Ceph-Bio-400 angle task
- Category
- medical-imaging-angle-estimation
- Direction
- Higher is better
- Sample
- 960
- Source type
- independent-evaluator
- Evaluator
- MedVision-V0 public leaderboard
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact MedVision-V0 leaderboard row GLM-4.6V-Flash (9B); corresponding public model repository https://huggingface.co/zai-org/GLM-4.6V-Flash. Scores are from the evaluator's current public Tables 2-4. The project's README says the leaderboard uses MedVision v1.0.0 annotations; it separately describes an open-weight limit of 1,000 samples per subtask and rounded totals in the public tables. No immutable evaluated weights revision, prompt, decoding setting, or reasoning mode is attached to the row.
- Context
- The source publishes rounded aggregate table denominators for detection and size, not an exact per-model scored N; no sampleCount is assigned for those cells. Distance and angle totals are explicit. No confidence interval is reported. The leaderboard documentation contains broader dataset-version statements, while its specific benchmark note says all leaderboard numbers use v1.0.0 annotations; the table does not bind an immutable model checkpoint or inference settings.
NutriBench UPR: Energy (kcal) (Show measurement, test conditions, and source)
- Source value
- 25.5
- Score
- 25.5
- Metric
- Unusable Prediction Rate (UPR)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
NutriBench UPR: Protein (g) (Show measurement, test conditions, and source)
- Source value
- 25.2
- Score
- 25.2
- Metric
- Unusable Prediction Rate (UPR)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
NutriBench UPR: Carb. (g) (Show measurement, test conditions, and source)
- Source value
- 25.4
- Score
- 25.4
- Metric
- Unusable Prediction Rate (UPR)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
NutriBench UPR: Total Fat (g) (Show measurement, test conditions, and source)
- Source value
- 26.8
- Score
- 26.8
- Metric
- Unusable Prediction Rate (UPR)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
NutriBench SMAPE_adj: Energy (kcal) (Show measurement, test conditions, and source)
- Source value
- 75
- Score
- 75
- Metric
- abstention-adjusted symmetric mean absolute percentage error (SMAPE_adj)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
NutriBench SMAPE_adj: Protein (g) (Show measurement, test conditions, and source)
- Source value
- 82.5
- Score
- 82.5
- Metric
- abstention-adjusted symmetric mean absolute percentage error (SMAPE_adj)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
NutriBench SMAPE_adj: Carb. (g) (Show measurement, test conditions, and source)
- Source value
- 81.2
- Score
- 81.2
- Metric
- abstention-adjusted symmetric mean absolute percentage error (SMAPE_adj)
- Unit
- %
- Benchmark version
- NutriBench (arXiv:2606.08948v1, Table 1 in Appendix Section 7.2)
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. GLM-4.6V-Flash (9B as labeled in the paper), released-form general open-weight MLLM baseline, not the separately fine-tuned NutriMLLM variant; text-only NutriBench evaluation using the common structured Prompt 2; Table 1 reports nutrient-specific UPR and abstention-adjusted SMAPE_adj. The paper does not report the exact evaluation sample size, checkpoint revision, quantization, or inference runtime.
- Context
- NutriBench comprises 11,857 meal descriptions, but the paper does not state the exact evaluation subset size for this table; sampleCount is omitted. No confidence interval is reported. Table 1 marks UPR and SMAPE_adj in percent with lower values better. The paper defines SMAPE_adj as abstention-adjusted symmetric mean absolute percentage error and states a 0-200% range.
Page 1 of 6
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
GLM-4.6V-FlashNemotron 3 Nano 45Llama 4 Maverick 45MiMo V2.6 Flash 45Qwen 3 VL 8B 45
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as GLM-4.6V-Flash, chosen by model line, developer, size, price and release date.
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about GLM-4.6V-Flash
How large is the context window of GLM-4.6V-Flash?
128,000 tokens. That is how much text GLM-4.6V-Flash reads at once, about 190 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
GLM-4.6V-FlashRank 98 of 103128K
DeepSeek-V3.2128K
DeepSeek-R1128K
Can I run GLM-4.6V-Flash locally?
Yes, the weights are open.
Stated by Z.ai
- Open weights
- yes
- License
- MIT
- Parameters
- 9B
- active per token
- n/a
Sources and data date
Show all 20 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 27, 2026. |
| Output | Documented modalities. Retrieved October 27, 2026. |
| Research status | Research date October 15, 2026. 12 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | extractbench-run-date-and-unit Note The independent benchmark widget reports exact scores and generation setup, but does not state task snapshot/run date or score units. |
| Unresolved | extractbench-run-date-and-score-unit Note The exact card/result show already recorded ExtractBench values but do not resolve the missing task run date or score unit. No unit or date is inferred. |
| Not published | extractbench-dataset-revision-and-long-split-sample-count Note The official dataset page documents a corpus of 370 documents, but the pinned model-card result does not pin the dataset revision or publish the Long-split document count and scored denominator. The corpus size is not treated as the model run sample count. |
| Not published | immutable-checkpoint-and-inference-settings Note The leaderboard names the model family/size but does not bind scores to a weights digest, prompt, decoding settings, or reasoning mode. |
| Not published | rounded-denominators Note Detection and size table headers show rounded K sample totals; no exact per-model denominator is published in those cells. |
| Additional source | Hugging Face (retrieved October 15, 2026) · GLM-4.6V-Flash model card · Editorial description reviewed October 2, 2026 |
| Additional source | LlamaIndex ExtractBench official leaderboard (retrieved October 15, 2026; October 16, 2026) |
| Additional source | Pinned GLM-4.6V-Flash card ExtractBench result (retrieved October 16, 2026) |
| Additional source | MedVision-V0 public leaderboard (retrieved October 17, 2026) |
| Additional source | NutriMLLM: Multimodal Large Language Models for Dietary Micronutrient Analysis (arXiv:2606.08948v1), Table 1 in Appendix Section 7.2 (retrieved October 17, 2026) |
| Additional source | KorMedMCQA-V: A Multimodal Benchmark for Evaluating Vision-Language Models on the Korean Medical Licensing Examination (retrieved October 19, 2026) |
| Additional source | VGenST-Bench authors, arXiv:2605.22570v1 (retrieved October 20, 2026) |
| Additional source | oMLX Community Benchmarks (retrieved October 22, 2026) |
| Additional source | llm-bench.io (retrieved October 23, 2026) |
| Additional source | Model metadata (retrieved October 16, 2026)12 |
| Additional source | Model metadata (retrieved October 17, 2026) |
| Additional source | GLM-4.6V-Flash model card text · Editorial description reviewed October 2, 2026 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |