Kimi model family
Kimi K3.1 Pro Abliterated.
An abliterated variant of Kimi K3.1 Pro for reasoning, code and multi-step work. Its original model is an open-weight, multimodal release post-trained from Kimi K3.
AbliteratedOfficial model reference & HydraLabs internal testing
How we evaluated
Kimi K3.1 Pro Abliterated is a separate model from Neurion Pro and Neurion Flash. The official reference results below describe the original Kimi K3.1 Pro release.
- Original context window
- 1Mtokens
- Original total parameters
- 2.8TMixture-of-Experts
- Original active parameters
- 104Bper token
HydraLabs reports an approximately 2% aggregate benchmark decrease in internal testing of the Abliterated variant.
This aggregate estimate does not adjust individual official scores. The internal test suite, aggregation method and raw runs have not been published; the estimate is not independently verified.
Official benchmark values remain unchanged. Neurion Pro comparisons retain their existing HydraLabs-reported setup. Different harnesses, tool access and task sets can limit direct comparison.
How to read the resultsView the benchmark reportCoding Bench
DeepSWE v1.1
Notes & sources
The six original comparison columns reproduce Moonshot AI’s published Kimi K3.1 Pro results. These are reference scores for the original model, not measured scores for the Abliterated variant. Neurion Pro is a separate HydraLabs-reported comparison, retained only for an explicitly matching benchmark identity. Harnesses, budgets and task sets can differ. Neurion Flash is not included. Missing results remain —, never zero.
Sampling defaults: reasoning max, temperature 1.0, top-p 0.95; individual source protocols may override these. Elo is not a percentage. Asterisks on HLE values denote text-only subsets. Tool-assisted and unassisted results are separate.
Neurion Pro: HydraLabs-reported internal evaluation · Ultra · mini-swe-agent. Its evaluation protocol is separate from the publisher comparison.
Original Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonHydraLabs-reported internal evaluation · Ultra · mini-swe-agent
HydraLabs-reported Neurion Pro resultAgentic Bench
BrowseComp
Notes & sources
The six original comparison columns reproduce Moonshot AI’s published Kimi K3.1 Pro results. These are reference scores for the original model, not measured scores for the Abliterated variant. Neurion Pro is a separate HydraLabs-reported comparison, retained only for an explicitly matching benchmark identity. Harnesses, budgets and task sets can differ. Neurion Flash is not included. Missing results remain —, never zero.
Sampling defaults: reasoning max, temperature 1.0, top-p 0.95; individual source protocols may override these. Elo is not a percentage. Asterisks on HLE values denote text-only subsets. Tool-assisted and unassisted results are separate.
No exactly mapped Neurion Pro result is supplied for this benchmark.
No result supplied for: Claude Fable 5.1, DeepSeek-V4.1-Flash-1017, GLM-5.3, Neurion Pro.
Original Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonReasoning Bench
GPQA-Diamond
Notes & sources
The six original comparison columns reproduce Moonshot AI’s published Kimi K3.1 Pro results. These are reference scores for the original model, not measured scores for the Abliterated variant. Neurion Pro is a separate HydraLabs-reported comparison, retained only for an explicitly matching benchmark identity. Harnesses, budgets and task sets can differ. Neurion Flash is not included. Missing results remain —, never zero.
Sampling defaults: reasoning max, temperature 1.0, top-p 0.95; individual source protocols may override these. Elo is not a percentage. Asterisks on HLE values denote text-only subsets. Tool-assisted and unassisted results are separate.
Neurion Pro: HydraLabs-reported internal evaluation · Ultra · MelissaCode · Epoch scope supplied by HydraLabs. Its evaluation protocol is separate from the publisher comparison.
Original Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonHydraLabs-reported internal evaluation · Ultra · MelissaCode · Epoch scope supplied by HydraLabs
HydraLabs-reported Neurion Pro resultCyber Bench
ExploitBench
Notes & sources
The six original comparison columns reproduce Moonshot AI’s published Kimi K3.1 Pro results. These are reference scores for the original model, not measured scores for the Abliterated variant. Neurion Pro is a separate HydraLabs-reported comparison, retained only for an explicitly matching benchmark identity. Harnesses, budgets and task sets can differ. Neurion Flash is not included. Missing results remain —, never zero.
Sampling defaults: reasoning max, temperature 1.0, top-p 0.95; individual source protocols may override these. Elo is not a percentage. Asterisks on HLE values denote text-only subsets. Tool-assisted and unassisted results are separate.
No exactly mapped Neurion Pro result is supplied for this benchmark.
No result supplied for: DeepSeek-V4.1-Flash-1017, Neurion Pro.
Original Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. The source column includes fallback. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonOriginal Kimi K3.1 Pro publisher comparison; max reasoning unless noted. See the source card for the exact benchmark protocol.
Moonshot AI · official comparisonModel specifications.
Reference architecture and modalities of the original Kimi K3.1 Pro release.
- Architecture
- Mixture-of-Experts (MoE)
- Total / active parameters
- 2.8T / 104B
- Context length
- 1,048,576 tokens
- Modalities
- Text and image
- Layers / dense layers
- 93 / 1
- Attention layers
- 69 KDA + 24 Gated DSA
- Attention hidden dimension
- 7,168
- Attention heads
- 96
- DSA index head dimension
- 128
- DSA index heads / top-K
- 32 / 2,048
- Latent MoE dimension
- 3,584
- Expert hidden dimension
- 3,072
- Routed / shared experts
- 896 / 2
- Selected experts per token
- 16
- Vocabulary
- 160K
- Activation
- SiTU-GLU
- Vision encoder
- MoonViT-V2 · 401M parameters
- Quantization
- MXFP4 weights / MXFP8 activations
The full benchmark report.
Coding, agentic, reasoning and cybersecurity results.
All 47 official rows remain unchanged. Neurion Pro is shown separately where an exact benchmark match exists; Neurion Flash is not included.
| Benchmark | Kimi K3.1 Pro | Claude Fable 5.1 | GPT-6 Astra | Claude Opus 5.5 | DeepSeek-V4.1-Flash-1017 | GLM-5.3 | Neurion Pro |
|---|---|---|---|---|---|---|---|
| Terminal-Bench 4.0 | 65.7 | 57.9 | 58.2 | 66.4 | 42.3 | 41.8 | 53.2% — Ultra |
| Terminal-Bench-Science 0.1 | 38.1 | 40.0 | 68.1 | 63.3 | 15.7 | 8.1 | — |
| SWE Marathon v1.1 | 51.7 | 45.6 | 42.5 | 46.3 | 40.0 | 42.5 | — |
| BenchCAD w/o tools | 72.8 | 60.6 | — | 73.0 | 41.2 | — | — |
| BenchCAD w/ tools | 92.7 | 92.6 | 95.9 | 96.2 | 81.6 | — | — |
| Program Bench | 31.5 | 53.5 | 50.0 | 65.0 | 29.1 | 19.0 | — |
| DeepSWE v1.1 | 75.5 | 67.4 | 74.1 | 74.2 | 77.0 | 66.9 | 70.3% — Ultra |
| PostTrain Bench v1.2 | 36.5 | 44.6 | 41.9 | 43.8 | 30.5 | 34.5 | — |
| Kimi Code Bench 2.0 (Internal) | 77.1 | — | — | — | 76.1 | 61.7 | — |
| Kimi Code Bench 3.0 Long-horizon (Internal) | 61.2 | 58.8 | 66.5 | 58.5 | 51.5 | 45.5 | — |
| HLE-Full w/ tools | 70.0 | 65.6 | 57.2 | 67.7 | 58.8 | 62.5 | — |
| BrowseComp | 92.1 | — | 91.5 | 90.8 | — | — | — |
| BioMysteryBench | 80.6 | 81.6 | 78.9 | 81.9 | 80.0 | 79.6 | — |
| CharXiv (RQ) w/ python | 93.4 | 92.7 | 92.8 | 93.8 | 85.6 | — | — |
| MMMU-Pro w/ python | 90.0 | 88.4 | 87.8 | 89.8 | 82.5 | — | — |
| GDP.pdf w/ python | 85.7 | 83.8 | 85.5 | 83.9 | 73.0 | — | — |
| ChartoGraphy w/ python | 90.2 | 88.4 | 91.0 | 89.0 | 83.0 | — | — |
| BabyVision w/ python | 95.4 | 95.1 | 96.6 | 97.7 | 92.5 | — | — |
| ZeroBench_main w/ python (pass@5) | 63.0 | 59.0 | 58.0 | 61.0 | 51.0 | — | — |
| Agents' Last Exam (ALL) | 43.2 | — | 41.8 | 42.6 | 37.7 | 26.2 | — |
| Agents' Last Exam (CLI) | 44.2 | — | 44.2 | 40.7 | 36.0 | 29.4 | — |
| Automation Bench-AA | 73.3 | 59.4 | 68.5 | 69.5 | 66.6 | 62.2 | — |
| GDPval-AA v2.1 (Elo-score) | 1846 | 1754 | 1573 | 1866 | 1746 | 1650 | — |
| AA-Briefcase v1.1 (Elo-score) | 1678 | 1672 | 1569 | 1803 | 1570 | 1507 | — |
| APEX-Agents | 69.4 | 68.6 | 64.7 | 73.5 | 69.8 | 56.6 | — |
| OfficeQA Pro v2 | 64.4 | 54.5 | 54.4 | 62.2 | 40.4 | 31.1 | — |
| OfficeQA Pro | 67.4 | 68.6 | 66.2 | — | 62.4 | 58.6 | — |
| Frontier Finance | 58.6 | 60.8 | 45.0 | 56.7 | 54.5 | 57.5 | — |
| Harvey Legal Agent | 23.7 | 19.1 | — | — | 21.0 | — | — |
| HANDBOOK.md | 44.6 | 38.8 | — | 41.5 | 28.9 | 20.8 | — |
| OSWorld V2.1 | 79.6 | 80.7 | 75.8 | 81.8 | 69.3 | — | — |
| Business Operation (Internal) | 37.8 | 30.7 | 32.3 | — | 8.9 | 18.6 | — |
| DECK-Arena (Internal, Elo-score) | 1456 | — | 1423 | 1253 | 1231 | 1119 | — |
| Design Bench (Internal, Elo-score) | 957 | — | 1076 | 927 | 863 | 718 | — |
| Kimi Hallu Bench Hard w/ tools (Internal, lower is better) | 8.3 | 23.3 | 5.5 | 15.0 | 23.3 | 22.6 | — |
| GDP.pdf | 34.6 | 27.6 | 34.2 | 30.6 | 17.2 | 11.2 | — |
| HLE-Full | 51.1 | 57.6 | 53.6 | 59.8 | 42.6 | 43.0* | — |
| HLE-Diamond | 39.4 | 55.1 | 65.8 | 60.9 | 21.6 | 25.1* | — |
| GPQA-Diamond | 94.2 | 93.7 | 96.1 | 93.9 | 92.2 | 91.7 | 94.8% — Ultra |
| CharXiv (RQ) | 89.0 | 88.8 | 90.0 | 92.1 | 77.0 | — | — |
| MMMU-Pro | 84.1 | 86.3 | 86.9 | 87.7 | 80.6 | — | — |
| ZeroBench_main (avg@5 / pass@5) | 29.8 / 39.0 | 23.0 / 35.0 | 43.2 / 52.0 | 35.0 / 43.0 | 11.8 / 23.0 | — | — |
| PerceptionBench | 68.3 | 63.0 | 77.6 | 72.1 | 47.7 | — | — |
| Kimi Hallu Bench w/o tools (Internal, lower is better) | 5.4 | 11.4 | 4.2 | 6.8 | 13.1 | 14.8 | — |
| ExploitBench | 74.7 | 83.0 | 100 | 91.0 | — | 54.4 | — |
| ExploitGym | 20.7 | 30.4 | 42.4 | 34.5 | — | 15.0 | — |
| xdev-bench-hard (Internal) | 81.0 | — | — | — | 85.7 | 57.1 | — |
Evaluation notes & sources
The six original comparison columns reproduce Moonshot AI’s published Kimi K3.1 Pro results. These are reference scores for the original model, not measured scores for the Abliterated variant. Neurion Pro is a separate HydraLabs-reported comparison, retained only for an explicitly matching benchmark identity. Harnesses, budgets and task sets can differ. Neurion Flash is not included. Missing results remain —, never zero.
Max reasoning, temperature 1.0 and top-p 0.95 are the source defaults. Comparative Claude and GPT columns include fallback. The source card contains benchmark-specific protocols, task subsets and tool settings.
Composite ZeroBench avg@5 / pass@5 values remain unchanged in the table and use separate plots. Elo scores, lower-is-better results and text-only asterisks retain their original meaning.
HydraLabs reports an approximately 2% aggregate benchmark decrease for the Abliterated variant. This is a separate internal estimate, not an adjustment to any official cell; raw runs and the aggregation method have not been published.
Official Kimi K3.1 Pro model cardKimi K3.1 Pro technical report
Source checked: .