NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB has 74.043 billion total parameters. The selected IQ2 mixed (QAT) GGUF file is 22.018 GB.
Data snapshot: . 3 published tests in this catalog.
Model file and details
Provider
nvidia
Total parameters
74,042,720,848
Exact file size
22,018,326,528 bytes
Quantization / precision
IQ2 mixed (QAT)
Weight repository
nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF
Original model repository
nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16
Weight revision
724e3fee41becc4048998e2204748147876ed56b
Model revision
bca0a8cb74200a2eb7ca56eb8428b821c829bd6e
Runtime notes
Confirm model architecture and quantization support in your runner.
Recorded release date
2026-10-15
Date evidence
NVIDIA released the OpenMDW-1.1 Nemotron 3.5 Bolt 74B-A9B 24GB GGUF on 15 October 2026. Pinned Hugging Face metadata identifies 74,042,720,848 total stored parameters and the complete single-file language checkpoint at 22,018,326,528 bytes. The separate BF16 vision projector is 1,733,692,736 bytes and is not included in the selected language-weight size. NVIDIA documents 74B total / 9B active parameters, a 256K maximum context, mixed IQ2_XXS, NVFP4 and Q8_0 weight storage produced with quantization-aware distillation, and exact llama.cpp recipes for 28GB and 32GB discrete-GPU deployments plus 64GB and 128GB RTX/DGX Spark systems. These are documented recipes, not a RAM-only compatibility guarantee. NVIDIA measured 83.99 tokens/s with the optional LiLiCorr drafter on AIPerf SPEED-Bench-Coding using llama.cpp on a 128GB Surface Laptop Ultra running Windows 11 ARM x64; this hardware- and recipe-specific result is not used in quality ranking.
These scores were compiled from the linked sources. These published scores were not independently tested here. Different prompts, precision, thinking budgets and tools can change results.
GPQA Diamond
86.43 / 100
Hard science questions. This is a science reasoning test, not an overall intelligence score.
Benchmark efficiency: 3.925 published score points per GB of selected weights. This ratio is not an inference-speed measure; check the test settings to see whether the score covers the selected quantization.
Test settings: NVIDIA-reported GPQA Diamond percentage for the exact released 24GB mixed-precision IQ2 checkpoint. NVIDIA's technical report identifies this as the quantized deployment configuration; evaluations other than MCP-Atlas, WebArena and BrowseComp reuse the Nemotron 3.5 Super VL parent setup.
Publisher-run release result, not independent validation. The report does not fully enumerate prompts, shots, answer extraction, reasoning budget or run variance for this row. It does establish that 86.43 applies to the released 2-bit checkpoint rather than the BF16 Bolt model or a third-party conversion. The result is a quality measurement, not an inference-speed or hardware-fit claim.
Writing code to solve scientific problems. Published results can depend on the test scaffold.
Benchmark efficiency: 1.863 published score points per GB of selected weights. This ratio is not an inference-speed measure; check the test settings to see whether the score covers the selected quantization.
Test settings: NVIDIA-reported SciCode percentage for the exact released 24GB mixed-precision IQ2 checkpoint. The technical report separates the 41.02 quantized result from 44.49 for the BF16 Bolt checkpoint and says this evaluation reuses the Nemotron 3.5 Super VL parent setup.
Publisher-run release result, not an independent test. Scientific-coding scaffolds, tool access, generation budget and harness details can materially affect SciCode, and the report does not fully enumerate them for this row.
Agentic terminal-use task success. Publisher-reported percentages are retained, but agent scaffold, tools, timeouts and reasoning budget can materially affect results.
Benchmark efficiency: 2.315 published score points per GB of selected weights. This ratio is not an inference-speed measure; check the test settings to see whether the score covers the selected quantization.
Test settings: NVIDIA-reported Terminal-Bench 2.1 percentage for the exact released 24GB mixed-precision IQ2 checkpoint. The technical report separates the 50.98 quantized result from 53.93 for the BF16 Bolt checkpoint and says this evaluation reuses the Nemotron 3.5 Super VL parent setup.
Publisher-run release result, not independent validation. Agent scaffold, tools, timeouts, environment and reasoning budget materially affect Terminal-Bench, and the report does not fully enumerate those conditions for this row. The separate 83.99 tokens/s result used a specific RTX Spark system, llama.cpp and LiLiCorr recipe and is not inferred from this benchmark.
Context and version evidence: The repository and file name identify the selected model. Revision hashes are shown above when recorded. Tested context length, full hardware configuration and run date are not consistently recorded in this snapshot. Only settings explicitly stated in each benchmark record are known.
The 22.018 GB figure covers model weights in this file. It excludes the context cache, runtime, operating system and any separate multimodal files. It is not a measured RAM requirement or a token-generation speed.
Need a computer for NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB?
These complete configurations pass the initial weight-size check for the selected 22.02 GB file. Allow extra memory for context and software. Check the model’s runtime notes and your inference software before buying.
Listing evidence: 2026-09-25. Amazon.com base price visible in an owner-supplied screenshot reviewed 25 Sep 2026; capture time not supplied. Dated reference, not a live quote. Excludes card signup offers, coupons, tax and shipping.
Listing evidence: 2026-10-02. Apple US launch price for the base M5 Max Mac Studio, published 22 September 2026: USD 2,499. Dated manufacturer reference, not an Amazon offer; tax and shipping may differ.
Listing evidence: 2026-09-24. GMKtec, EVO-X2 / 64GB / 1TB, recorded 22 Sep 2026; exact ASIN capacity identified 24 Sep. This is a dated reference, not a current Amazon price; tax, shipping and bundle contents may differ.
Memory matches are estimates, not tested compatibility or speed. As an Amazon Associate I earn from qualifying purchases.
Confirm model architecture and quantization support in your runner. The selected weights use 22.018 GB; this excludes context cache, runtime buffers and other programs.
These are the first listed tiers that pass our reserve policy, not measured minimum requirements. System RAM and VRAM remain separate. Use the memory tool to change the reserve for your context and runner.