Inkling 975B
Thinking Machines
Parameters
975.0B total / 41.0B active
MoE: total / active
Architecture
MoE Transformer (per Ripple-65B Preview card comparison table)
Released
—
License
—
About
Comparison-anchor stub record created during the ricursiveAI/Ripple-65B-Preview ingest. The Ripple-65B Preview model card compares its instruct model against this model; parameters taken verbatim from that comparison table (975B total, 41B active, 45T pretraining tokens). Full ingestion pending - needs a dedicated HuggingFace-page ingest.
Benchmark Scores
| Benchmark | Score | Date | Standing |
|---|---|---|---|
|
GPQA Diamond
stem_reasoning
|
86.58%
|
— |
#39/80
better than 52%
|
|
SciCode
|
76.38%
|
— |
#4/27
better than 88%
|
|
Terminal Bench 2.1
coding_agent
|
70.29%
|
— |
#26/44
better than 42%
|
|
AutomationBench (Public)
general_agent
|
12.33%
|
— |
#10/11
better than 10%
|
|
SWE-bench Verified
coding_agent
|
88.53%
|
— |
#15/60
better than 76%
|
|
DeepSWE 1.1
coding_agent
|
2.08%
|
— |
#33/34
better than 3%
|
|
SWE-bench Pro
coding_agent
|
60.54%
|
— |
#36/77
better than 53%
|
|
SWE-bench Pro v2.0 Hard
coding_agent
|
18.39%
|
— |
#4/5
better than 25%
|
|
AIME 2025
stem_reasoning
|
88.21%
|
— |
#8/24
better than 65%
|
|
AIME 26
stem_reasoning
|
97.32%
|
— |
#8/45
better than 84%
|
|
AA-LCR
long_context
|
93.21%
|
— |
#5/42
better than 90%
|
|
HMMT Feb 26
stem_reasoning
|
85.23%
|
— |
#19/44
better than 58%
|
|
StrongREJECT v2 (adaptive)
safety
|
100.00%
|
— |
#1/5
better than 100%
|
|
FORTRESS Benign
safety
|
100.00%
|
— |
#1/5
better than 100%
|
|
TauBench V3 Banking
general_agent
|
52.10%
|
— |
#7/17
better than 62%
|
|
MMLU
knowledge
|
96.06%
|
— |
#5/16
better than 73%
|
|
AIME 2024
stem_reasoning
|
93.10%
|
— |
#2/7
better than 83%
|
|
Terminal-Bench 2.1 (Terminus-2)
coding_agent
|
51.92%
|
— |
#30/44
better than 33%
|
|
MCP-Atlas
general_agent
|
82.77%
|
— |
#13/54
better than 75%
|
|
MMLU-Pro
knowledge
|
88.06%
|
— |
#9/47
better than 83%
|
|
IFBench
instruction_following
|
92.85%
|
— |
#6/42
better than 88%
|
What is a good GPQA Diamond score? What is a good SciCode score? What is a good Terminal Bench 2.1 score?
Hardware Requirements
Estimated VRAM per quantization at 4,096 tokens context — MoE: all experts must fit in VRAM, only 41.0B are active per token
| Quantization | Weights | KV Cache | Total VRAM | On your hardware |
|---|---|---|---|---|
| FP16 — 16-bit | 1950.0 GB | 151.3 GB | 2198.8 GB | Which GPUs can run it? Calculate on your GPU |
| INT8 — 8-bit | 975.0 GB | 75.7 GB | 1099.4 GB | Which GPUs can run it? Calculate on your GPU |
| INT4 — 4-bit | 487.5 GB | 37.8 GB | 549.7 GB | Which GPUs can run it? Calculate on your GPU |
| FP4 — 4-bit | 487.5 GB | 37.8 GB | 549.7 GB | Which GPUs can run it? Calculate on your GPU |
Only datacenter GPUs — FP16 needs 2198.8 GB. For consumer hardware, choose the INT4 build with a multi-GPU setup.
No comments yet
Be the first to share your thoughts!