RTX 4090
Used-card condition, power supply and case clearance need separate checks.
MODEL / NEX AGI
Nex-N2.5-mini
Model card checked 2026-10-25 · Memory fit estimated; speed unknown unless a separately linked measurement matches your configuration.
Also searched as: nex-n2.5-mini
35B checkpoint (the repository's safetensors total 35.1B parameters). Q4 budget estimated as 35 × 0.5 × 1.2 = 21 GB before KV cache. The publisher's card documents only SGLang: its Nex-N2.5-mini recipe is one node with 2× H100 (--tp 2) on the customized fork image nexagi/sglang:v0.5.18-nex-patch, with --reasoning-parser qwen3 and --tool-call-parser qwen3_coder for the family's reasoning_effort modes (none / medium / high). The card publishes no vLLM recipe, so this record lists SGLang as the only publisher-documented runtime. The native window is 262,144 tokens in config.json and the mini recipe sets no explicit --context-length, unlike the Max recipe. Verify the fork image, conversion, tool parser and reasoning parser before deployment; the card's linked GitHub repository also did not resolve when checked on 2026-10-25; speed is unknown.
Primary sourceCheck this model against your hardwarePick a device or GPU to see memory fit, runtime path and the deployment template.
Total parameters35B
Estimated memory~21 GBQ4 estimated budget
Context262,144 (config max_position_embeddings)Text + Image
LicenseApache-2.0Multilingual
01 / BEST FOR
- agentic workflows
- computer use
- multimodal agents
02 / ESTIMATED FIT
03 / DEPLOY WITH
These are runtime examples, not validated installation recipes. Fit estimates assume quantized files; official repositories may hold full-precision weights. MLX, GGUF and TensorRT require matching conversions or engines.
SGLangAdvanced
python3 -m sglang.launch_server --model-path nex-agi/Nex-N2.5-mini --port 30000Watch out
35B checkpoint (the repository's safetensors total 35.1B parameters). Q4 budget estimated as 35 × 0.5 × 1.2 = 21 GB before KV cache. The publisher's card documents only SGLang: its Nex-N2.5-mini recipe is one node with 2× H100 (--tp 2) on the customized fork image nexagi/sglang:v0.5.18-nex-patch, with --reasoning-parser qwen3 and --tool-call-parser qwen3_coder for the family's reasoning_effort modes (none / medium / high). The card publishes no vLLM recipe, so this record lists SGLang as the only publisher-documented runtime. The native window is 262,144 tokens in config.json and the mini recipe sets no explicit --context-length, unlike the Max recipe. Verify the fork image, conversion, tool parser and reasoning parser before deployment; the card's linked GitHub repository also did not resolve when checked on 2026-10-25; speed is unknown.
Your purchase shortlist
Filter by a runtime path and an installed-capacity floor you already established for your workload. This does not calculate required VRAM. Inputs stay in this browser.
RTX 5090
Check Blackwell kernels for the exact quantization; more VRAM does not certify faster decode.
Radeon AI PRO R9700
Verify OS, driver and ROCm/Vulkan support. Experimental forks are not stock-runtime support.
Framework Desktop Ryzen AI Max+ 395 · 128GB
BIOS GPU allocation and system use reduce available memory; 128 GB RAM is not 128 GB dedicated VRAM.
Mac Studio M5 Ultra · 96GB
m5 ultra 96g and m5 ultra 96gb refer to this capacity. Memory is shared with macOS; verify shipping date and MLX conversion.
DGX Spark
ARM64 packages and GB10 kernels must support the chosen model. Capacity is not a latency result.
Checked model specification
| Issuer / checkpoint | Values | Source / verified |
|---|---|---|
| Nex AGI / nex-agi/Nex-N2.5-mini | 35B total; 262,144 (config max_position_embeddings); Apache-2.0 | Publisher model card / 2026-10-25 |
Applicability and failure modes
Inference experiments only. The Q4 memory budget is estimated, not a tested maximum context. Unsupported kernels, missing vision projectors, incorrect tool parsers and longer-context KV allocation can fail even when weights fit. No successful deployment, quality or speed is guaranteed.
Compare hardware purchase paths