ROG Strix SCAR 18 (2026) · Core Ultra 9 290HX Plus · RTX 5090 Laptop 24 GB · 32 GB
ASUS · ROG Strix SCAR 18 (2026)
639 text models run well · 896 GB/s bandwidth
Models on this hardware
Start here
File Q4_K_M · 11.6 GB · unsloth/gpt-oss-20b-GGUF · What is Q4?
Memory Needs ~16 GB · ~18 GB on ROG Strix SCAR 18 (2026) · RTX 5090 Laptop · 24 GB VRAM, including ~2 GB the system keeps
ollama run hf.co/unsloth/gpt-oss-20b-GGUF:Q4_K_M- Install Ollama
- Run the command above
- Ask:
Explain what a context window is, in two sentences.
File UD-Q4_K_M · 16.9 GB · unsloth/gemma-4-26B-A4B-it-GGUF · What is Q4?
Memory Needs ~19 GB · ~21 GB on ROG Strix SCAR 18 (2026) · RTX 5090 Laptop · 24 GB VRAM, including ~2 GB the system keeps
ollama run hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_M- Install Ollama
- Run the command above
- Ask:
Explain what a context window is, in two sentences.
Text
Text · LiveBench · overall
0–100 · higher is better
- Color = publisher
Source: LiveBench 2026-06-25 · collected 2026-10-18 · Open models without a score are not drawn (no score ≠ 0).
| Smaug MiniQ4 · published params · source | 76.9 | Not on the LMArena text leaderboard of 2026-10-22 | ~23 GBTight fit: Needs ~23 GB: runs well, about 1 GB left beyond the system reserve | Q4 · published params |
|---|---|---|---|---|
| Qwen3.8 27BQ4 · published params · source | 75.3 | 1,437 ± 5 | ~23 GBTight fit: Needs ~23 GB: runs well, about 1 GB left beyond the system reserve | Q4 · published params |
| Qwen3.6 27BQ4 · published params · source | 64.0 | Not on the LMArena text leaderboard of 2026-10-22 | ~23 GBTight fit: Needs ~23 GB: runs well, about 1 GB left beyond the system reserve | Q4 · published params |
| Qwen3 14BQ4 · published params · source | No comparable published result | Not on the LMArena text leaderboard of 2026-10-22 | ~14 GB | Q4 · published params |
| Qwen3.5 9BQ4 · published params · source | No comparable published result | Not on the LMArena text leaderboard of 2026-10-22 | ~9 GB | Q4 · published params |
| Qwen3 8BQ4 · published params · source | No comparable published result | Not on the LMArena text leaderboard of 2026-10-22 | ~9 GB | Q4 · published params |
| gpt-oss-20bQ4 · published params · source | No comparable published result | 1,317 ± 6 | ~18 GB | Q4 · published params |
| Gemma 4 26B A4BQ4 · published params · source | No comparable published result | 1,437 ± 7 | ~21 GB | Q4 · published params |
LMArena ratings in this table: LMArena 2026-10-22 · collected 2026-10-25 · dataset · CC BY 4.0
Tight fit: it runs, but leaves less than 2.4 GB free beyond the 2 GB the system keeps (the memory figure already includes that reserve).
Every reviewed text model here 37
Every one that runs here, well or slowly, scored or not (28 run well, 9 slowly); 20 of 37 have a LiveBench or LMArena result. Run well first, then tight fits, then slow; within each, by Hub downloads over 30 days, read 2026-11-19, not by quality.
| Qwen3 14B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~14 GB | 9.0 | Apache 2.0 | 13.2M |
|---|---|---|---|---|---|---|
| Qwen3.5 9B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~9 GB | 5.7 | Apache 2.0 | 6.5M |
| Qwen3 8B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~9 GB | 5.0 | Apache 2.0 | 6.1M |
| gpt-oss-20b | Not in LiveBench 2026-06-25 | 1,317 | ~18 GB | 11.6 | Apache 2.0 | 5.6M |
| Gemma 4 26B A4B | Not in LiveBench 2026-06-25 | 1,437 | ~21 GB | 16.9 | Apache 2.0 | 5M |
| Gemma 4 E4B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~8 GB | 5.0 | Apache 2.0 | 4.4M |
| Llama 3.1 8B Instruct | Not in LiveBench 2026-06-25 | 1,211 | ~8 GB | 4.9 | Llama 3.1 | 4.1M |
| Gemma 4 E2B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~6 GB | 3.1 | Apache 2.0 | 2.8M |
| Qwen3 4B Instruct 2507 | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~5 GB | 2.5 | Apache 2.0 | 2.8M |
| Qwen3 1.7B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~4 GB | 1.1 | Apache 2.0 | 2.7M |
| Gemma 4 12B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~11 GB | 7.1 | Apache 2.0 | 1.6M |
| Llama 3.2 1B Instruct | Not in LiveBench 2026-06-25 | 1,111 | ~3 GB | 0.8 | Llama 3.2 | 1.3M |
| Llama 3.2 3B Instruct | Not in LiveBench 2026-06-25 | 1,167 | ~5 GB | 2.0 | Llama 3.2 | 823.8K |
| DeepSeek R1 0528 Qwen3 8B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~9 GB | 5.0 | MIT | 523.8K |
| Phi-4 mini | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~5 GB | 2.5 | MIT | 301.1K |
| Ministral 3 14B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~13 GB | 8.2 | Apache 2.0 | 207.6K |
| Phi-4 | Not in LiveBench 2026-06-25 | 1,256 | ~13 GB | 8.0 | MIT | 199.5K |
| SmolLM3 3B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~5 GB | 1.9 | Apache 2.0 | 185.3K |
| Olmo 3 7B Instruct | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~8 GB | 4.5 | Apache 2.0 | 133.1K |
| Mistral Small 3.2 24B | Not in LiveBench 2026-06-25 | 1,356 | ~21 GB | 14.3 | Apache 2.0 | 109.9K |
| Ministral 3 8B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~9 GB | 5.2 | Apache 2.0 | 90.7K |
| Ministral 3 3B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~5 GB | 2.1 | Apache 2.0 | 72.5K |
| Granite 4.2 8B | Not in LiveBench 2026-06-25 | 1,290 | ~8 GB | 5.3 | Apache 2.0 | 59.5K |
| Granite 4.2 3B | Not in LiveBench 2026-06-25 | 1,290 | ~5 GB | 2.2 | Apache 2.0 | 40.8K |
| Devstral Small 2 24B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~20 GB | 14.3 | Apache 2.0 | 24.8K |
| Qwen3.8 27B | 75.3 | 1,437 | ~23 GBTight fit | 16.5 | Apache 2.0 | 6.9M |
| Qwen3.6 27B | 64.0 | Not on the LMArena text leaderboard of 2026-10-22 | ~23 GBTight fit | 16.8 | Apache 2.0 | 2.5M |
| Smaug Mini | 76.9 | Not on the LMArena text leaderboard of 2026-10-22 | ~23 GBTight fit | 16.5 | Apache 2.0 | 1.1K |
| Gemma 4 31B | Not in LiveBench 2026-06-25 | 1,452 | ~26 GBRuns slowly | 18.3 | Apache 2.0 | 6.3M |
| Qwen3 Coder 30B A3B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | ~25 GBRuns slowly | 18.6 | Apache 2.0 | 1.9M |
| Qwen3.5 35B A3B | Not in LiveBench 2026-06-25 | 1,394 | ~29 GBRuns slowly | 22.0 | Apache 2.0 | 1.4M |
| Nemotron 3 Nano 30B A3B | Not in LiveBench 2026-06-25 | 1,314 | ~32 GBRuns slowly | 24.6 | NVIDIA Nemotron Open Model | 1.3M |
| GLM-4.7 Flash | Not in LiveBench 2026-06-25 | 1,365 | ~25 GBRuns slowly | 18.3 | MIT | 1.2M |
| Nemotron 3.5 Lightning 30B A3B | 44.5 | Arena rated the NVFP4 build, NVIDIA-Nemotron-3.5-Lightning-30B-A3B (NVFP4); this entry is the BF16 reference, which NVIDIA reports with different scores | ~25 GBRuns slowly | 18.9 | OpenMDW 1.1 | 974.1K |
| Qwen3 30B A3B Instruct 2507 | Not in LiveBench 2026-06-25 | 1,382 | ~25 GBRuns slowly | 18.6 | Apache 2.0 | 747.4K |
| Granite 4.2 30B | Not in LiveBench 2026-06-25 | 1,340 | ~25 GBRuns slowly | 17.7 | Apache 2.0 | 37K |
| Olmo 3 32B Think | Not in LiveBench 2026-06-25 | 1,306 | ~26 GBRuns slowly | 19.5 | Apache 2.0 | 11.8K |
Also runs slowly
Above the 24 GB of VRAM, the layers that do not fit run on the CPU from PC RAM, up to 48 GB in all (listed RAM: 32 GB). Planning estimate, not a speed figure. The 9 reviewed models that run slowly are in the table above.
- Qwen3-30B-A3BQwen~25 GBruns well with 32 GB
- Qwen3-30B-A3B-Thinking-2507Qwen~25 GBruns well with 32 GB
- Qwen3-30B-A3B-BaseQwen~25 GBruns well with 32 GB
- Tongyi-DeepResearch-30B-A3BAlibaba-NLP~25 GBruns well with 32 GB
- SciJudge-30BOpenMOSS-Team~25 GBruns well with 32 GB
- Rubicon-PreviewinclusionAI~25 GBruns well with 32 GB
- SciJudge-30B-2605OpenMOSS-Team~25 GBruns well with 32 GB
- SciThinker-30BOpenMOSS-Team~25 GBruns well with 32 GB
- browsesafeperplexity-ai~25 GBruns well with 32 GB
- tt-x15-megatron-51-30Blaion~25 GBruns well with 32 GB
- tt-x2_clip-hi0p2-60-30Blaion~25 GBruns well with 32 GB
- tt-x10-fsdp2-fa2-117-30Blaion~25 GBruns well with 32 GB
- tt-x2_clip-hi0p05-10-30Blaion~25 GBruns well with 32 GB
- tt-x4_temperature-t0p7-45-30Blaion~25 GBruns well with 32 GB
- tt-x1_lr-lr8e6-60-30Blaion~25 GBruns well with 32 GB
- tt-x16-entropy0p003-42-30Blaion~25 GBruns well with 32 GB
- tt-x2_clip-hi0p01-15-30Blaion~25 GBruns well with 32 GB
- tt-x1_lr-lr4e6-30-30Blaion~25 GBruns well with 32 GB
- tt-x4_temperature-t1p0-45-30Blaion~25 GBruns well with 32 GB
- tt-x0_verifier-shaped-20-30Blaion~25 GBruns well with 32 GB
- tt-x0_verifier-binary-20-30Blaion~25 GBruns well with 32 GB
- tt-x5_gradnorm-gn1p8laion~25 GBruns well with 32 GB
- tt-x5_gradnorm-gn0p9laion~25 GBruns well with 32 GB
- tt-x3_kl-kl0laion~25 GBruns well with 32 GB
- Qwen3-32BQwen~27 GBruns well with 32 GB
- Qwen2.5-32B-InstructQwen~27 GBruns well with 32 GB
- Qwen2.5-Coder-32B-InstructQwen~27 GBruns well with 32 GB
- DeepSeek-R1-Distill-Qwen-32Bdeepseek-ai~27 GBruns well with 32 GB
- Qwen1.5-32B-ChatQwen~27 GBruns well with 32 GB
- QwQ-32BQwen~27 GBruns well with 32 GB
- Qwen1.5-32BQwen~27 GBruns well with 32 GB
- QwQ-32B-PreviewQwen~27 GBruns well with 32 GB
- OLMo-2-0325-32B-Instructallenai~27 GBruns well with 32 GB
- OLMo-2-0325-32Ballenai~27 GBruns well with 32 GB
- Qwen2.5-Coder-32BQwen~27 GBruns well with 32 GB
- Qwen3-Nemotron-32B-GenRM-Principlenvidia~27 GBruns well with 32 GB
- SWE-Dev-32Bzai-org~27 GBruns well with 32 GB
- StepFun-Prover-Preview-32Bstepfun-ai~27 GBruns well with 32 GB
- Skywork-SWE-32BSkywork~27 GBruns well with 32 GB
- Qwen3-Nemotron-32B-RLBFFnvidia~27 GBruns well with 32 GB
- NFT-32Bnvidia~27 GBruns well with 32 GB
- Qwen3-32B-AWorldinclusionAI~27 GBruns well with 32 GB
- OpenReasoning-Nemotron-32Bnvidia~27 GBruns well with 32 GB
- SynLogic-32BMiniMaxAI~27 GBruns well with 32 GB
- Nemotron-Terminal-32Bnvidia~27 GBruns well with 32 GB
- OpenMath-Nemotron-32Bnvidia~27 GBruns well with 32 GB
- OpenCodeReasoning-Nemotron-32Bnvidia~27 GBruns well with 32 GB
- xLAM-2-32b-fc-rSalesforce~27 GBruns well with 32 GB
- WebDancer-32BAlibaba-NLP~27 GBruns well with 32 GB
- SynLogic-Mix-3-32BMiniMaxAI~27 GBruns well with 32 GB
- OpenCodeReasoning-Nemotron-1.1-32Bnvidia~27 GBruns well with 32 GB
- StepFun-Formalizer-32Bstepfun-ai~27 GBruns well with 32 GB
- NextCoder-32Bmicrosoft~27 GBruns well with 32 GB
- OpenCodeReasoning-Nemotron-32B-IOInvidia~27 GBruns well with 32 GB
- OREAL-32Binternlm~27 GBruns well with 32 GB
- DRIVE-RLtencent~27 GBruns well with 32 GB
- DRIVE-SFTtencent~27 GBruns well with 32 GB
- OProver-32Bm-a-p~27 GBruns well with 32 GB
- Qwen3-32B-ABCOpenMOSS-Team~27 GBruns well with 32 GB
- OProver-32B-Basem-a-p~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_warmup-ratio_0-05_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_adam-beta1_0-93_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_lr_1e-5_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_adam-beta1_0-95_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_warmup-ratio_0-01_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_global-batch-size_64_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_Qwen3-32Blaion~27 GBruns well with 32 GB
- a2-rl-e2egit_large-50-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_adam-beta1_0-91_Qwen3-32Blaion~27 GBruns well with 32 GB
- sft__g1_gptlong_top8_train_on_obs_fa3__40-0__Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_num-train-epochs_4.0_Qwen3-32Blaion~27 GBruns well with 32 GB
- nemotron-terminal-corpus-unified-10000__Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_num-train-epochs_7.0_Qwen3-32Blaion~27 GBruns well with 32 GB
- OProver-32B-Round1m-a-p~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_adam-beta1_0-97_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_global-batch-size_32_Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-a1mfc-dev1-maxeps-swes-r2eg-32b__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-a1mfc-gfistaqc-scaff-maxeps-swes-r2eg-32b__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemotron-terminal-corpus-unified-31600__Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_learning-rate_1e-06_Qwen3-32Blaion~27 GBruns well with 32 GB
- GLM-4.6-stackexchange-overflow-sandboxes-32eps-65k-reasoning_num-train-epochs_8-0_Qwen3-32Blaion~27 GBruns well with 32 GB
- Qwen3-32B-NL2Bash-31steplaion~27 GBruns well with 32 GB
- syh-r2eg-askl-glm_4-7_trac_jupi_-gfi-swes-rand-filt-10K_glm_4-7_trac_jupi_32Blaion~27 GBruns well with 32 GB
- nemotron-terminal-corpus-unified-316__Qwen3-32Blaion~27 GBruns well with 32 GB
- sft_GLM-4-7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k_Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-a1mfc-dev1-maxeps-32b__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-a1unix-a1mfc-gfistaqc-scaff-maxeps-swes-r2eg-32b__Qwen3-32Blaion~27 GBruns well with 32 GB
- sft_r2egym-nl2bash-stackoverflow-inferredbugs-32B_Qwen3-32Blaion~27 GBruns well with 32 GB
- g1_selective_top8_diverse_31600_glm47_traces_32Blaion~27 GBruns well with 32 GB
- Qwen3-32B-SweSmith-20steplaion~27 GBruns well with 32 GB
- tasrep-a1mfc-gfistaqc-dev1-scaff-maxeps-swes-r2eg-32b__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-a1mfc-gfistaqc-dev1-scaff-maxeps-swes-r2eg-32b-10pct__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemotron-terminal-corpus-unified-3160__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemotron-terminal-corpus-unified-1000__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-a1mfc-gfistaqc-dev1-scaff-maxeps-swes-r2eg-32b-3pct__Qwen3-32Blaion~27 GBruns well with 32 GB
- nemosci-tasrep-nemodebug-a1mfc-gfistaqc-scaff-maxeps-swes-r2eg-32b__Qwen3-32Blaion~27 GBruns well with 32 GB
- deepseek-coder-33b-instructdeepseek-ai~28 GBruns well with 32 GB
- Smaug-34B-v0.1abacusai~28 GBruns well with 32 GB
- deepseek-coder-33b-basedeepseek-ai~28 GBruns well with 32 GB
- CodeLlama-34b-Instruct-hfmeta-llama~28 GBruns well with 32 GB
- OpenMath-CodeLlama-34b-Python-hfnvidia~28 GBruns well with 32 GB
- MetaMath-Bagel-DPO-34Babacusai~28 GBruns well with 32 GB
- CodeLlama-34b-hfmeta-llama~28 GBruns well with 32 GB
- MetaMath-bagel-34b-v0.2-c1500abacusai~28 GBruns well with 32 GB
- MM-Orc-Vic-bagel-34b-c1000abacusai~28 GBruns well with 32 GB
- MM-OV-bagel-DPO-34b-c1000-250abacusai~28 GBruns well with 32 GB
- layerskip-codellama-34Bfacebook~28 GBruns well with 32 GB
- Kun-LabelModelm-a-p~28 GBruns well with 32 GB
- CodeLlama-34b-Python-hfmeta-llama~28 GBruns well with 32 GB
- Smaug-Mixtral-v0.1abacusai~38 GBQ4 · tensor count
- xLAM-8x7b-rSalesforce~38 GBQ4 · tensor count
- NeKo-v0-post-correctionnvidia~38 GBQ4 · tensor count
- xLAM-v0.1-rSalesforce~38 GBQ4 · tensor count
- Qwen2-57B-A14B-InstructQwen~46 GBQ4 · tensor count
- Qwen2-57B-A14BQwen~46 GBQ4 · tensor count
Adjust the estimate: quantization, context, reserve
Vision
Vision · LMArena
rating · higher is better
- Color = publisher
Source: LMArena 2026-10-19 · collected 2026-10-25 · dataset · CC BY 4.0 · Models without a rating are left out, not scored as zero.
| Qwen3.8 27BQ4 · published params · source | 1,242 ± 8 | ~23 GBTight fit: Needs ~23 GB: runs well, about 1 GB left beyond the system reserve | Q4 · published params |
|---|---|---|---|
| Gemma 4 26B A4BQ4 · published params · source | 1,242 ± 6 | ~21 GB | Q4 · published params |
| gemma-3-27b-itQ4 · Hugging Face count · source | 1,160 ± 8 | ~23 GBTight fit: Needs ~23 GB: runs well, about 1 GB left beyond the system reserve | Q4 · Hugging Face count |
| Molmo2-8BQ4 · Hugging Face count · source | 1,112 ± 20 | ~9 GB | Q4 · Hugging Face count |
| Qwen2-VL-7B-InstructQ4 · Hugging Face count · source | 1,031 ± 11 | ~9 GB | Q4 · Hugging Face count |
| Molmo-7B-D-0924Q4 · Hugging Face count · source | 994 ± 14 | ~9 GB | Q4 · Hugging Face count |
| Llama-3.2-11B-Vision-InstructQ4 · Hugging Face count · source | 993 ± 11 | ~11 GB | Q4 · Hugging Face count |
| MiniCPM-V-2_6Q4 · Hugging Face count · source | 962 ± 16 | ~9 GB | Q4 · Hugging Face count |
LMArena ratings in this table: LMArena 2026-10-19 · collected 2026-10-25 · dataset · CC BY 4.0
Tight fit: it runs, but leaves less than 2.4 GB free beyond the 2 GB the system keeps (the memory figure already includes that reserve).
Also runs slowly
Above the 24 GB of VRAM, the layers that do not fit run on the CPU from PC RAM, up to 48 GB in all (listed RAM: 32 GB). Planning estimate, not a speed figure.
- Gemma 4 31BGoogle~26 GBruns well with 32 GB
- Qwen3.5 35B A3BAlibaba~29 GBruns well with 32 GB
- ERNIE-4.5-VL-28B-A3B-Base-Paddlebaidu~25 GBruns well with 32 GB
- ERNIE-4.5-VL-28B-A3B-Base-PTbaidu~25 GBruns well with 32 GB
- ERNIE-4.5-VL-28B-A3B-Paddlebaidu~25 GBruns well with 32 GB
- ERNIE-4.5-VL-28B-A3B-PTbaidu~25 GBruns well with 32 GB
- ERNIE-4.5-VL-28B-A3B-Thinkingbaidu~25 GBruns well with 32 GB
- Hy-Embodied-VLM-1.0tencent~25 GBruns well with 32 GB
- gemma-4-31Bgoogle~26 GBruns well with 32 GB
- Holo2-30B-A3BHcompany~26 GBruns well with 32 GB
- InternVL3_5-30B-A3BOpenGVLab~26 GBruns well with 32 GB
- InternVL3_5-30B-A3B-FlashOpenGVLab~26 GBruns well with 32 GB
- InternVL3_5-30B-A3B-HFOpenGVLab~26 GBruns well with 32 GB
- InternVL3_5-30B-A3B-InstructOpenGVLab~26 GBruns well with 32 GB
- InternVL3_5-30B-A3B-MPOOpenGVLab~26 GBruns well with 32 GB
- InternVL3_5-30B-A3B-PretrainedOpenGVLab~26 GBruns well with 32 GB
- Ising-Calibration-1.5-31B-BF16nvidia~26 GBruns well with 32 GB
- Keye-VL-2.0-30B-A3BKwai-Keye~26 GBruns well with 32 GB
- Prompt-RefineHiDream-ai~26 GBruns well with 32 GB
- Qwen3-VL-30B-A3B-InstructQwen~26 GBruns well with 32 GB
- Qwen3-VL-30B-A3B-ThinkingQwen~26 GBruns well with 32 GB
- UI-Venus-1.5-30B-A3BinclusionAI~26 GBruns well with 32 GB
- aya-vision-32bCohereLabs~27 GBruns well with 32 GB
- Holotron-3-NanoHcompany~27 GBruns well with 32 GB
- Holotron4-30B-A3BHcompany~27 GBruns well with 32 GB
- LLaVA-NeXT-Video-32B-Qwenlmms-lab~27 GBruns well with 32 GB
- AesCode-32Bmicrosoft~28 GBruns well with 32 GB
- chameleon-30bfacebook~28 GBruns well with 32 GB
- Cosmos-Reason2-32Bnvidia~28 GBruns well with 32 GB
- GTA1-32BSalesforce~28 GBruns well with 32 GB
- OriOn-Qwen-SR1lightonai~28 GBruns well with 32 GB
- Qwen2.5-VL-32B-InstructQwen~28 GBruns well with 32 GB
- Qwen3-VL-32B-InstructQwen~28 GBruns well with 32 GB
- Qwen3-VL-32B-ThinkingQwen~28 GBruns well with 32 GB
- ScaleCUA-32BOpenGVLab~28 GBruns well with 32 GB
- AdvancedMathBench-AutoVerifierinternlm~29 GBruns well with 32 GB
- fibo-scene-analyzerbriaai~29 GBruns well with 32 GB
- Holo-3.1-35B-A3BHcompany~29 GBruns well with 32 GB
- Holo3-35B-A3BHcompany~29 GBruns well with 32 GB
- Holo4-35B-A3BHcompany~29 GBruns well with 32 GB
- Intern-S2-Mobiusinternlm~29 GBruns well with 32 GB
- LLaVA-NeXT-Video-34B-DPOlmms-lab~29 GBruns well with 32 GB
- LLaVA-NeXT-Video-34B-DPO-hfllava-hf~29 GBruns well with 32 GB
- LLaVA-NeXT-Video-34B-hfllava-hf~29 GBruns well with 32 GB
- llava-v1.6-34b-hfllava-hf~29 GBruns well with 32 GB
- MiniCPM-V-4.7-35B-A3Bopenbmb~29 GBruns well with 32 GB
- Qwen3.5-35B-A3B-BaseQwen~29 GBruns well with 32 GB
- Qwen3.6-35B-A3BQwen~29 GBruns well with 32 GB
- Intern-S2-Previewinternlm~30 GBruns well with 32 GB
- InternVL2_5-38BOpenGVLab~31 GBruns well with 32 GB
- InternVL2_5-38B-MPOOpenGVLab~31 GBruns well with 32 GB
- InternVL3_5-38BOpenGVLab~31 GBruns well with 32 GB
- InternVL3_5-38B-HFOpenGVLab~31 GBruns well with 32 GB
- InternVL3_5-38B-InstructOpenGVLab~31 GBruns well with 32 GB
- InternVL3_5-38B-MPOOpenGVLab~31 GBruns well with 32 GB
- InternVL3_5-38B-PretrainedOpenGVLab~31 GBruns well with 32 GB
- InternVL3-38BOpenGVLab~31 GBruns well with 32 GB
- InternVL3-38B-hfOpenGVLab~31 GBruns well with 32 GB
- InternVL3-38B-InstructOpenGVLab~31 GBruns well with 32 GB
- InternVL3-38B-PretrainedOpenGVLab~31 GBruns well with 32 GB
- Skywork-R1V-38BSkywork~31 GBruns well with 32 GB
- Skywork-R1V2-38BSkywork~31 GBruns well with 32 GB
- Skywork-R1V3-38BSkywork~31 GBruns well with 32 GB
- SkyworkVL-38BSkywork~31 GBruns well with 32 GB
- InternVL3_5-38B-FlashOpenGVLab~32 GBruns well with 32 GB
- InternVL-Chat-V1-2OpenGVLab~33 GBQ4 · Hugging Face count
- InternVL-Chat-V1-2-PlusOpenGVLab~33 GBQ4 · Hugging Face count
- InternVL2-40BOpenGVLab~33 GBQ4 · Hugging Face count
With 32 GB you also run… 1
The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.
- mmE5-mllama-11b-instructintfloat~28 GB16-bit est. · Hugging Face count
471 more vision models in the catalog have no usable memory figure or parameter count for this profile. Browse Vision models
Image
Image · Prompt alignment · GenEval
0–1 · higher is better
Source: GenEval · 2023 · reviewed 2026-10-05 · Published quality in the original benchmark preset; not measured on this hardware.
| Stable Diffusion XL 1.0Platform path documented · stable-diffusion.cpp · source | 0.55 | Not on the LMArena text-to-image leaderboard of 2026-10-07 | Documented path without a memory figure | Platform path documented |
|---|---|---|---|---|
| Stable Diffusion 1.5Platform path documented · stable-diffusion.cpp · source | 0.43 | Not on the LMArena text-to-image leaderboard of 2026-10-07 | Documented path without a memory figure | Platform path documented |
| ERNIE ImagePublisher minimum · CUDA · source | No comparable published result | Not on the LMArena text-to-image leaderboard of 2026-10-07 | ~24 GB | Publisher minimum |
| ERNIE Image TurboPublisher minimum · CUDA · source | No comparable published result | Not on the LMArena text-to-image leaderboard of 2026-10-07 | ~24 GB | Publisher minimum |
| ETCHR-FLUX.2-klein-9B16-bit est. · params · source | No comparable published result | Not reviewed for LMArena | ~24 GB | 16-bit est. · params |
| FLUX.2 [dev]Conditional path · CUDA · source | No comparable published result | 1,146 ± 4 | ~24 GB | Conditional path |
| HiDream-O1-Image-Dev-260416-bit est. · params · source | No comparable published result | Not reviewed for LMArena | ~24 GB | 16-bit est. · params |
| HunyuanImage-2.1Conditional path · CUDA · source | No comparable published result | Not on the LMArena text-to-image leaderboard of 2026-10-07 | ~24 GB | Conditional path |
LMArena ratings in this table: LMArena 2026-10-07 · collected 2026-10-25 · dataset · CC BY 4.0
With 32 GB you also run… 16
The next memory size on this ladder. Figures are the requirement on this hardware type.
- Ideogram 4.0 FastFal~25 GB16-bit est. · params
- Ideogram 4.0 Fast (Quality)Fal~25 GB16-bit est. · params
- Ideogram 4.0 InstantFal~25 GB16-bit est. · params
- Qwen-Image-2.1-PE-T2IQwen~25 GB16-bit est. · params
- FLUX.2 [klein] 9BBlack Forest Labs~29 GBPublisher minimum
- FLUX.2 [klein] Base 9BBlack Forest Labs~29 GBPublisher minimum
- FLUX.2-klein-9b-kvblack-forest-labs~29 GBPublisher minimum
- FLUX.1 Krea [dev]Black Forest Labs~31 GB16-bit est. · params
- FLUX.1-Canny-devblack-forest-labs~31 GB16-bit est. · params
- FLUX.1-Depth-devblack-forest-labs~31 GB16-bit est. · params
- FLUX.1-Kontext-devblack-forest-labs~31 GB16-bit est. · params
- SRPOTencent~31 GB16-bit est. · params
- Qwen-Image-NVFP4nvidia~32 GB16-bit est. · params
- Step1X-Edit-v1p1-diffusersstepfun-ai~32 GB16-bit est. · params
- Step1X-Edit-v1p2stepfun-ai~32 GB16-bit est. · params
- Step1X-Edit-v1p2-previewstepfun-ai~32 GB16-bit est. · params
231 more image models in the catalog have no usable memory figure or parameter count for this profile. Browse Image models
Video
| Wan2.2-TI2V-5BConditional path · CUDA · source | ~24 GB | Conditional path |
|---|---|---|
| Wan2.2-TI2V-5B-DiffusersConditional path · CUDA · source | ~24 GB | Conditional path |
| ContentV-8B16-bit est. · params · source | ~22 GB | 16-bit est. · params |
| HunyuanVideo-1.516-bit est. · params · source | ~22 GB | 16-bit est. · params |
| CogVideoX-5b-I2V16-bit est. · params · source | ~16 GB | 16-bit est. · params |
| CogVideoX1.5-5B16-bit est. · params · source | ~16 GB | 16-bit est. · params |
| CogVideoX1.5-5B-I2V16-bit est. · params · source | ~16 GB | 16-bit est. · params |
| SkyReels-A116-bit est. · params · source | ~16 GB | 16-bit est. · params |
How this list works Documented execution paths on this hardware first, then publisher minimums (CUDA or Apple MPS).
With 32 GB you also run… 5
The next memory size on this ladder. Figures are the requirement on this hardware type.
- mochi-1-previewgenmo~26 GB16-bit est. · params
- Open-Sora-v2hpcai-tech~29 GB16-bit est. · params
- Video-As-Prompt-CogVideoX-5BByteDance~29 GB16-bit est. · params
- LTX-2.5 FastLightricks~32 GBPublisher minimum
- LTX-2.5 ProLightricks~32 GBPublisher minimum
124 more video models in the catalog have no usable memory figure or parameter count for this profile. Browse Video models
Audio
Audio · Word error rate · Open ASR
% · lower is better
Source: Word error rate · Open ASR · reviewed 2026-10-20 · Published quality in the original benchmark preset; not measured on this hardware.
| Whisper large-v3-turboPlatform path documented · whisper.cpp / CUDA · source | 7.83 | Documented path without a memory figure | Platform path documented |
|---|---|---|---|
| Whisper large-v3Platform path documented · whisper.cpp / CUDA · source | 7.44 | Documented path without a memory figure | Platform path documented |
| Fish Audio S2 ProPublisher minimum · CUDA · source | No comparable published result | ~24 GB | Publisher minimum |
| higgs-audio-v3-8b-stt16-bit est. · params · source | No comparable published result | ~24 GB | 16-bit est. · params |
| higgs-audio-v3-8b-stt-v216-bit est. · params · source | No comparable published result | ~24 GB | 16-bit est. · params |
| Ming-3.0-asr-tiny16-bit est. · params · source | No comparable published result | ~24 GB | 16-bit est. · params |
| Ming-3.0-audio-tiny16-bit est. · params · source | No comparable published result | ~24 GB | 16-bit est. · params |
| Ming-3.0-tts-tiny16-bit est. · params · source | No comparable published result | ~24 GB | 16-bit est. · params |
With 32 GB you also run… 10
The next memory size on this ladder. Figures are the requirement on this hardware type.
- AuK-Flashtencent~25 GBPublisher minimum
- AuKtencent~25 GBPublisher minimum
- Kimi-Audio-7Bmoonshotai~26 GB16-bit est. · params
- Kimi-Audio-7B-Instructmoonshotai~26 GB16-bit est. · params
- MERaLiON-3-10BMERaLiON~26 GB16-bit est. · params
- MERaLiON-AudioLLM-Whisper-SEA-LIONMERaLiON~26 GB16-bit est. · params
- MERaLiON-2-10BMERaLiON~27 GB16-bit est. · params
- MERaLiON-2-10B-ASRMERaLiON~27 GB16-bit est. · params
- acestep-captionerACE-Step~28 GB16-bit est. · params
- acestep-transcriberACE-Step~28 GB16-bit est. · params
1,984 more audio models in the catalog have no usable memory figure or parameter count for this profile. Browse Audio models
Embeddings
Embeddings · MTEB (Multilingual, v2) · mean(task)
0–100 · higher is better
- Color = publisher
| Qwen3-Embedding 8B8-bit · published params · source | 70.6 | ~12 GB | 8-bit · published params |
|---|---|---|---|
| Qwen3-Embedding 4B8-bit · published params · source | 69.5 | ~7 GB | 8-bit · published params |
| Harrier OSS v1 (0.6B)8-bit · published params · source | 69.0 | ~3 GB | 8-bit · published params |
| Qwen3-Embedding 0.6B8-bit · published params · source | 64.3 | ~3 GB | 8-bit · published params |
| Multilingual E5 Large Instruct8-bit · published params · source | 63.2 | ~3 GB | 8-bit · published params |
| EmbeddingGemma 28-bit · published params · source | 61.4 | ~3 GB | 8-bit · published params |
| EmbeddingGemma (300M)8-bit · published params · source | 61.1 | ~3 GB | 8-bit · published params |
| BGE-M38-bit · published params · source | 59.6 | ~3 GB | 8-bit · published params |
Runs well within 24 GB of VRAM · planning estimates, not measured speed
Measurements
Published tokens per second, exactly as measured; never estimated.
| Llama 2 7B Q4_0 | 158.14 ± 4.70 | Q4_0 | CUDA | llama.cpp CUDA scoreboard · community result |
|---|---|---|---|---|
| Llama 2 7B Q4_0 | 154.31 ± 1.61 | Q4_0 | Vulkan | llama.cpp Vulkan scoreboard · community result |
Specifications
31 sourced fields; the source is named once per group, and a field from another source links to it. Official specifications
Markers: the best LiveBench-scored text model that runs well here, and the lightest text model that does not run well here but runs well with 32 GB. Q4 planning estimates, not measured fits.
Reviewed · Manufacturer source
- CPU cores
- 24
- Processor
- Intel Core Ultra 9 Processor 290HX Plus
- Graphics processor
- NVIDIA GeForce RTX 5090 Laptop GPU
- GPU runtime status
- Not stated by the source
- Memory capacity
- 32 GB
- Memory kind
- System RAM
- Battery capacity
- 90 Wh
- CPU threads
- 24
- Dimensions (cm)
- 39.9 x 29.8 x 2.35 ~ 3.50
- Display size
- 18 in
- Display resolution
- 3840 x 2400
- Form factor
- laptop (18-inch)
- GPU VRAM
- 24 GB
- GPU memory type
- GDDR7
- GPU power
- 175 W
- Installed storage
- 1TB PCIe 4.0 NVMe M.2 SSD
- Maximum supported memory
- 128 GB
- Memory data rate
- 6,400 MT/s
- Memory slots
- 2
- System memory type
- DDR5
- NPU performance
- 13 TOPS · INT8
- Operating system
- Windows 11 Home
- Ports
- 2x Thunderbolt 5 (DisplayPort / power delivery / G-SYNC) · 3x USB 3.2 Gen 2 Type-A · 1x HDMI 2.1 FRL · 1x 2.5G LAN · 1x 3.5mm combo audio
- Power supply
- 450 W
- Product SKU
- G835LXG-DB96
- Installed memory configuration
- DDR5-6400 SO-DIMM (2 SO-DIMM slots, max 128GB)
- Weight
- 3.73 kg
Intel Core Ultra 9 Processor 290HX Plus · NVIDIA GeForce RTX 5090 Laptop GPU with 32 GB installed system RAM and 24 GB dedicated GPU VRAM. Runtime support and local inference speed are not verified.
Photo: ASUS family photo · memory configuration is not pictured. · Manufacturer source
Memory bandwidth is not measured model speed; verify runtime and checkpoint compatibility. Hardware specification reviewed: 2026-10-18.
How estimates work
- Candidates
- Reviewed pages use published parameter counts; automatic text candidates use official repository tensor counts (packed quantized weights and unsupported architectures excluded).
- Planning
- Q4 weights + 20% allowance + reserve: 25% of unified or system memory (at least 2 GB, at most 8 GB; 2 GB here) or 2 GB per dedicated GPU. Adjustable in “Adjust the estimate”. No KV cache is calculated from context.
- Fit tiers
- Runs well: the estimate fits the fast memory (VRAM, or unified memory within the macOS GPU share). Runs slowly: within the machine's memory but outside the fast part; a graphics card offloads layers to PC RAM (the listed RAM, or 32 GB assumed). Doesn't run: above both. Tiers are planning estimates, never a speed.
- Vision
- Vision-language models use the Q4 planning estimate from the parameter count (published by the model page, or read from Hugging Face). Classic vision models (classify, detect, segment) use a 16-bit estimate. No vision benchmark is tracked, so rows carry no score.
- Media and embeddings
- Media rows use documented execution paths (exact hardware or platform) and publisher minimums; a published requirement above this memory moves a model to the next tier. Without either, a 16-bit estimate from the published parameter count (× 1.2 + 2 GB), labelled as such. Embeddings use 8-bit weights + 20% + 2 GB.
Local AI basics: parameters, Q4 and memory · Full model catalog · Data and methodology
Alternatives
Other machines on the same memory ladder (VRAM); each page lists the models that fit it.
