MapikaDecision modelOpen source
Decider 35B A3B NVFP4
Popularityhardly any signal
under 1out of 100same as last month
Summary
Decider 35B A3B NVFP4 is the quantized form of Decider 35B A3B. The readout stays the same. Attention projections and expert components are quantized to NVFP4; delta-net convolutions and input projections, routers, gates, embeddings, and the LM head remain in bf16. Published memory use therefore drops from 65 GB for the bf16 checkpoint to 19.6 GB. The checkpoint targets vLLM, TensorRT-LLM, and SGLang on Blackwell GPUs.
Across the published fixtures, NVFP4 scores 1.0 to 1.5 points below bf16.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 55 published benchmark results
Page 1 of 5
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.
TypeSafe workflow decisions (Show measurement, test conditions, and source)
- Source value
- 0.843
- Score
- 0.843
- Metric
- accuracy
- Unit
- ratio
- Category
- decision
- Direction
- Higher is better
- Sample
- 102
- Source type
- vendor-reported
- Evaluator
- Mapika Decider 35B NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exported NVFP4 weights served with vLLM 0.29 on one B300; 102 TypeSafe workflow rows; paired BF16 comparison.
In-task validation rows (Show measurement, test conditions, and source)
- Source value
- 0.882
- Score
- 0.882
- Metric
- accuracy
- Unit
- ratio
- Category
- decision
- Direction
- Higher is better
- Sample
- 847
- Source type
- vendor-reported
- Evaluator
- Mapika Decider 35B NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exported NVFP4 weights served with vLLM 0.29 on one B300; 847 in-task validation rows; paired BF16 comparison.
TypeSafe workflow decisions negative log-likelihood (Show measurement, test conditions, and source)
- Source value
- 0.365
- Score
- 0.365
- Metric
- negative log-likelihood (nats)
- Unit
- No unit provided
- Benchmark version
- Mapika/decider MODEL_CARD_35B_NVFP4.md; repository revision d6fc522aed7fc392576a9b66a330409bc6c382bd; exported NVFP4 checkpoint; ModelOpt 0.46.1
- Direction
- Lower is better
- Sample
- 102
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Decider-35B-A3B NVFP4 exported checkpoint derived from v1; ModelOpt 0.46.1; exported quantized checkpoint, not simulated fake quantization; card revision d6fc522aed7fc392576a9b66a330409bc6c382bd.
- Context
- These scores belong to the exported NVFP4 checkpoint. Simulated fake-quantization and BF16 results are excluded.
TypeSafe workflow decisions argmax agreement with BF16 reference (Show measurement, test conditions, and source)
- Source value
- 0.961
- Score
- 0.961
- Metric
- argmax agreement with BF16 reference
- Unit
- ratio
- Benchmark version
- Mapika/decider MODEL_CARD_35B_NVFP4.md; repository revision d6fc522aed7fc392576a9b66a330409bc6c382bd; exported NVFP4 checkpoint; ModelOpt 0.46.1
- Direction
- Higher is better
- Sample
- 102
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Decider-35B-A3B NVFP4 exported checkpoint derived from v1; ModelOpt 0.46.1; argmax agreement compared with the BF16 reference; exported quantized checkpoint, not simulated fake quantization; card revision d6fc522aed7fc392576a9b66a330409bc6c382bd.
- Context
- These scores belong to the exported NVFP4 checkpoint. Simulated fake-quantization and BF16 results are excluded.
TypeSafe workflow decisions mean total variation (Show measurement, test conditions, and source)
- Source value
- 0.051
- Score
- 0.051
- Metric
- mean total variation distance
- Unit
- ratio
- Benchmark version
- Mapika/decider MODEL_CARD_35B_NVFP4.md; repository revision d6fc522aed7fc392576a9b66a330409bc6c382bd; exported NVFP4 checkpoint; ModelOpt 0.46.1
- Direction
- Lower is better
- Sample
- 102
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Decider-35B-A3B NVFP4 exported checkpoint derived from v1; ModelOpt 0.46.1; exported quantized checkpoint, not simulated fake quantization; card revision d6fc522aed7fc392576a9b66a330409bc6c382bd.
- Context
- These scores belong to the exported NVFP4 checkpoint. Simulated fake-quantization and BF16 results are excluded.
In-task validation negative log-likelihood (Show measurement, test conditions, and source)
- Source value
- 0.339
- Score
- 0.339
- Metric
- negative log-likelihood (nats)
- Unit
- No unit provided
- Benchmark version
- Mapika/decider MODEL_CARD_35B_NVFP4.md; repository revision d6fc522aed7fc392576a9b66a330409bc6c382bd; exported NVFP4 checkpoint; ModelOpt 0.46.1
- Direction
- Lower is better
- Sample
- 847
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Decider-35B-A3B NVFP4 exported checkpoint derived from v1; ModelOpt 0.46.1; exported quantized checkpoint, not simulated fake quantization; card revision d6fc522aed7fc392576a9b66a330409bc6c382bd.
- Context
- These scores belong to the exported NVFP4 checkpoint. Simulated fake-quantization and BF16 results are excluded.
In-task validation argmax agreement with BF16 reference (Show measurement, test conditions, and source)
- Source value
- 0.967
- Score
- 0.967
- Metric
- argmax agreement with BF16 reference
- Unit
- ratio
- Benchmark version
- Mapika/decider MODEL_CARD_35B_NVFP4.md; repository revision d6fc522aed7fc392576a9b66a330409bc6c382bd; exported NVFP4 checkpoint; ModelOpt 0.46.1
- Direction
- Higher is better
- Sample
- 847
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Decider-35B-A3B NVFP4 exported checkpoint derived from v1; ModelOpt 0.46.1; argmax agreement compared with the BF16 reference; exported quantized checkpoint, not simulated fake quantization; card revision d6fc522aed7fc392576a9b66a330409bc6c382bd.
- Context
- These scores belong to the exported NVFP4 checkpoint. Simulated fake-quantization and BF16 results are excluded.
In-task validation mean total variation (Show measurement, test conditions, and source)
- Source value
- 0.03
- Score
- 0.03
- Metric
- mean total variation distance
- Unit
- ratio
- Benchmark version
- Mapika/decider MODEL_CARD_35B_NVFP4.md; repository revision d6fc522aed7fc392576a9b66a330409bc6c382bd; exported NVFP4 checkpoint; ModelOpt 0.46.1
- Direction
- Lower is better
- Sample
- 847
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider NVFP4 model card
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Decider-35B-A3B NVFP4 exported checkpoint derived from v1; ModelOpt 0.46.1; exported quantized checkpoint, not simulated fake quantization; card revision d6fc522aed7fc392576a9b66a330409bc6c382bd.
- Context
- These scores belong to the exported NVFP4 checkpoint. Simulated fake-quantization and BF16 results are excluded.
Decision Index 0.1, headline raw area-mean score (Show measurement, test conditions, and source)
- Source value
- 54.3
- Score
- 54.3
- Metric
- Decision Index 0.1 headline score
- Unit
- points
- Benchmark version
- Decision Index edition 0.1 (2026-09-22); Mapika results revision d6fc522aed7fc392576a9b66a330409bc6c382bd; Decider 35B A3B exported NVFP4 checkpoint derived from v1, ModelOpt 0.46.1
- Category
- decision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider results table (external leaderboard transcription)
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Mapika's external-leaderboard table reports decider-35b-a3b (NVFP4) in Decision Index edition 0.1, dated 2026-09-22. The edition method defines the headline as 100 times the raw mean of 19 benchmark areas. No model-specific N is published.
- Context
- Vendor-reported transcription: Mapika says these external-leaderboard numbers were read from the board and were not measured in its repository. The original board artifact was not captured here. No per-model N or CI is published. This is a source-specific index score, not an assigned rank or eligibility for any other index.
Decision Index 0.1, knowledge area raw score (Show measurement, test conditions, and source)
- Source value
- 0.51
- Score
- 0.51
- Metric
- Decision Index 0.1 raw area score
- Unit
- ratio
- Benchmark version
- Decision Index edition 0.1 (2026-09-22); Mapika results revision d6fc522aed7fc392576a9b66a330409bc6c382bd; Decider 35B A3B exported NVFP4 checkpoint derived from v1, ModelOpt 0.46.1
- Category
- decision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider results table (external leaderboard transcription)
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Mapika reports this per-area value for the immediately identified decider-35b-a3b (NVFP4) Decision Index 0.1 panel entry. The edition method defines area values as equal-weight means of raw benchmark scores. four reported constituents: GPQA, GSM8K, CRUXEval and MMLU. No per-model or per-area N is published.
- Context
- Vendor-reported transcription. The results table omits NVFP4 from the area-column header, but the immediately preceding paragraph introduces the same panel as the 35B NVFP4 entry. The original leaderboard was not captured here; no per-area N or CI is published.
Decision Index 0.1, language area raw score (Show measurement, test conditions, and source)
- Source value
- 0.61
- Score
- 0.61
- Metric
- Decision Index 0.1 raw area score
- Unit
- ratio
- Benchmark version
- Decision Index edition 0.1 (2026-09-22); Mapika results revision d6fc522aed7fc392576a9b66a330409bc6c382bd; Decider 35B A3B exported NVFP4 checkpoint derived from v1, ModelOpt 0.46.1
- Category
- decision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider results table (external leaderboard transcription)
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Mapika reports this per-area value for the immediately identified decider-35b-a3b (NVFP4) Decision Index 0.1 panel entry. The edition method defines area values as equal-weight means of raw benchmark scores. edition 0.1 raw language-area score. No per-model or per-area N is published.
- Context
- Vendor-reported transcription. The results table omits NVFP4 from the area-column header, but the immediately preceding paragraph introduces the same panel as the 35B NVFP4 entry. The original leaderboard was not captured here; no per-area N or CI is published.
Decision Index 0.1, retrieval area raw score (Show measurement, test conditions, and source)
- Source value
- 0.34
- Score
- 0.34
- Metric
- Decision Index 0.1 raw area score
- Unit
- ratio
- Benchmark version
- Decision Index edition 0.1 (2026-09-22); Mapika results revision d6fc522aed7fc392576a9b66a330409bc6c382bd; Decider 35B A3B exported NVFP4 checkpoint derived from v1, ModelOpt 0.46.1
- Category
- decision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Mapika, Decider results table (external leaderboard transcription)
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Mapika reports this per-area value for the immediately identified decider-35b-a3b (NVFP4) Decision Index 0.1 panel entry. The edition method defines area values as equal-weight means of raw benchmark scores. edition 0.1 raw retrieval-area score. No per-model or per-area N is published.
- Context
- Vendor-reported transcription. The results table omits NVFP4 from the area-column header, but the immediately preceding paragraph introduces the same panel as the 35B NVFP4 entry. The original leaderboard was not captured here; no per-area N or CI is published.
Page 1 of 5
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 37 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 6, 2026
Decider 35B A3B NVFP4d1-2B 0d1-Omni-2B 0GLiDE 0Laya 0
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.
Decider 35B A3B NVFP4 compared
- 1Decider 35B A3B NVFP4 vs. Decider 2B37 shared tests
- 2Decider 35B A3B NVFP4 vs. Clef-Flash37 shared tests
- 3Decider 35B A3B NVFP4 vs. CLM-v0.1-8B37 shared tests
- 4Decider 35B A3B NVFP4 vs. GLiNER2.5-Decide37 shared tests
- 5Decider 35B A3B NVFP4 vs. Intern-Decision-0.8B37 shared tests
- 6Decider 35B A3B NVFP4 vs. Intern-Decision-2B37 shared tests
- 7Decider 35B A3B NVFP4 vs. Intern-Decision-4B37 shared tests
- 8Decider 35B A3B NVFP4 vs. Julia 137 shared tests
- 9Decider 35B A3B NVFP4 vs. Kev 0.8B37 shared tests
Similar models
More models of the same kind as Decider 35B A3B NVFP4, 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
API
Further details
Sources and data date
Show all 16 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.