Qwen3.5 9B Instruct
Local OKExcellentA · 35
Best for: Quality, Coding, Reasoning·Pop: 86/100
Perf: est. 104 tok/s, typical error 5% · first token ~0.3s
Fits in 12 GB VRAM with room to spare. Best for quality, coding, reasoning on RTX 5070.
Best Local AI Models for RTX 5070 (12GB)
The RTX 5070 brings Blackwell architecture to the mid-range. On 8B models it reaches est. 98 tok/s, typical error 5%, 34% faster than the 4070 SUPER while keeping 12GB VRAM. Best for users prioritizing speed with 7B-9B models.
Quick answer
The best local LLM for the RTX 5070 is Qwen3.5 9B Instruct (est. 104 tok/s, typical error 5%) on its 12GB VRAM. It uses ~7GB of VRAM; the RTX 5070 handles up to 9B parameter models at Q4. A 14B model at Q4 runs at est. 60 tok/s, typical error 15%. (ModelFit, 2026)
Sizing rule: a Q4 model needs about 0.6 GB of VRAM per billion parameters, and ModelFit budgets 90% of this card's 12GB for weights, context, and KV-cache. The per-size table below uses that same budget. Other strong fits: Gemma 4 12B (12B, ~8GB) and Ornith 1.0 9B (9B, ~5.6GB). On this card Qwen3.5 9B Instruct runs at est. 104 tok/s, typical error 5%. What it will not run: a 20B MoE (3.6B active) model does not fit in VRAM alone, so it spills to system RAM over PCIe and slows sharply.
$ollama run qwen3.5:9b
- Top pick
- Qwen3.5 9B Instruct
- Decode speed
- est. 104 tok/s, typical error 5%
- VRAM needed
- ~7 GB
Speeds are ModelFit estimates from memory bandwidth and model size, not measured benchmarks. Typical error: the median gap between the ModelFit speed model and measured runs on hardware the model was not fitted on, for this backend and model size.
Cite this page: ModelFit, RTX 5070 12GB Local LLM (2026): Fastest 12GB, est. 98 tok/s, https://modelfit.io/gpu/rtx-5070/, updated October 2026, CC BY 4.0.
Specs
VRAM
12 GB GDDR7
Speed (8B Q4)
est. 98 tok/stypical error 5%
Bandwidth
672 GB/s
Architecture
Blackwell
Max model size
Up to 9B parameter models
Compatibility
10 excellent, 0 workable
Vendor
Table 1
RTX 5070 estimated tokens/sec by model size
Q4_K_M · ModelFit estimate| Model size | Est. speed | Fit on 12GB |
|---|---|---|
| 7B | est. 111 tok/stypical error 5% | Fits in VRAM |
| 14B | est. 60 tok/stypical error 15% | Fits in VRAM |
| 20B MoE (3.6B active) | est. 97 tok/stypical error 15% | CPU offload (slow) |
| 32B | est. 3 tok/stypical error 5% | CPU offload (slow) |
| 35B MoE (3B active) | est. 52 tok/stypical error 15% | CPU offload (slow) |
| 70B | est. 1 tok/stypical error 10% | CPU offload (slow) |
| 120B MoE (5.1B active) | est. 34 tok/stypical error 15% | CPU offload (slow) |
Fig. 1RTX 5070: estimated speed by model size
Decode speed, generic Q4_K_M model of each size at 4K context, 672 GB/s, typical error 5-15%
Fits in 12 GBCPU offload (slow)
7B111
14B60
20B MoE (3.6B active)97
32B3
35B MoE (3B active)52
70B1
120B MoE (5.1B active)34
tok/s est.ModelFit estimates, not measured benchmarks: decode speed of a generic Q4_K_M model of each size at 4K context on the RTX 5070's 672 GB/s bandwidth. MoE rows scale by active parameters (decode reads only the active experts), so a 35B MoE runs far faster than a dense 32B. "CPU offload" sizes exceed the 12GB VRAM; dense models slow to a crawl there, MoE models degrade less because hot experts stay GPU-resident. What typical error means.
Fig. 2
Context costs VRAM too.
Qwen3.5 9B Instruct loads ~7 GB of weights; at 16k context the KV cache adds ~0.5 GB (still fits the ~11 GB usable VRAM), and at 64k it adds ~2.0 GB (still fits).
Weights 7.0 GBKV cache · 16k context 0.5 GBHeadroom 3.3 GBRuntime reserve 1.2 GB
The model leaves about 3.3 GB of the usable VRAM budget free after weights and 16k context.
KV-cache figures assume an fp16 cache, the llama.cpp/Ollama default. Standard GQA models use a size-class estimate (8 KV heads x 128 head dim class); hybrid linear-attention models (Qwen3.5/3.6, Qwen3-Next) use the exact per-token cost from their published config, since only their sparse full-attention layers cache KV. A q8_0 KV cache roughly halves either figure. Estimates, not measurements.
Get told when a model beats Qwen3.5 9B Instruct on your RTX 5070
One email when a new open-weight model is a better fit for 12 GB of VRAM, plus the Thursday weekly.
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
Where to buy the RTX 5070
Storage & accessories for your model library
- Internal NVMe SSD · 2TBA Gen4 M.2 drive keeps your whole GGUF and quant collection on fast local storage, loading models straight off NVMe.Check price on Amazon
- USB4 NVMe Enclosure40Gbps external storage fast enough to run models from. Pair it with an M.2 drive for a portable model vault.Check price on Amazon
ModelFit may earn a commission on purchases through these links, at no extra cost to you.
RTX 5070 VRAM for AI: what actually fits?
12GB GDDR7 at 672 GB/s makes the RTX 5070 the fastest 12GB card for AI inference. Compared to the RTX 4070 SUPER (504 GB/s), bandwidth jumps 33%. This directly translates to faster token generation: est. 98 tok/s, typical error 5% vs est. 73 tok/s. The 12GB limit means 14B models still require quantization tricks. For 7B-9B workloads, this is the practical pick if you do not need the extra VRAM.
Table 2
RTX 5070 vs similar GPUs
| Hardware | Memory | Speed | Bandwidth |
|---|---|---|---|
| RTX 4070 SUPER | 12 GB | est. 73 tok/stypical error 5% | 504 GB/s |
| RTX 4070 | 12 GB | est. 73 tok/stypical error 5% | 504 GB/s |
| RTX 5070 | 12 GB | est. 98 tok/stypical error 5% | 672 GB/s |
| RTX 5070 Ti | 16 GB | est. 125 tok/stypical error 5% | 896 GB/s |
Recommended models
registry-verified10 modelsGemma 4 12B
Local OKOKB · 26
Best for: Chat, Coding, Multimodal·Pop: 80/100
Perf: est. 70 tok/s, typical error 15% · first token ~0.4s
Fits in 12 GB VRAM with room to spare. Best for chat, coding, multimodal on RTX 5070.
Ornith 1.0 9B
Local OKExcellentA · 48
Best for: Agentic coding on small machines·Pop: 76/100
Perf: est. 104 tok/s, typical error 5% · first token ~0.3s
Fits in 12 GB VRAM with room to spare. Best for agentic coding on small machines on RTX 5070.
Gemma 4 E4B (Q8)
Local OKExcellentA · 31
Best for: On-device, Mobile, Chat·Pop: 82/100
Perf: est. 105 tok/s, typical error 5% · first token ~1.5s
Fits in 12 GB VRAM with room to spare. Best for on-device, mobile, chat on RTX 5070.
LFM2.5 8B-A1B
Local OKExcellentA · 49
Best for: On-device agents, tool calling, multilingual chat·Pop: 72/100
Perf: est. 246 tok/s, typical error 15% · first token ~0.1s
Fits in 12 GB VRAM with room to spare. Best for on-device agents, tool calling, multilingual chat on RTX 5070.
MiMo-V2.6-Distill-Qwen-9B
Local OKExcellentA · 46
Best for: Agentic coding, Small machines·Pop: 60/100
Perf: est. 101 tok/s, typical error 5% · first token ~0.3s
Fits in 12 GB VRAM with room to spare. Best for agentic coding, small machines on RTX 5070.
Ministral 3 8B
Local OKExcellentA · 44
Best for: Chat, Coding, Vision·Pop: 74/100
Perf: est. 87 tok/s, typical error 5% · first token ~0.3s
Fits in 12 GB VRAM with room to spare. Best for chat, coding, vision on RTX 5070.
Ministral 3 14B
Local OKOKB · 16
Best for: Chat, Reasoning, Vision·Pop: 74/100
Perf: est. 61 tok/s, typical error 15% · first token ~0.5s
Fits in 12 GB VRAM with room to spare. Best for chat, reasoning, vision on RTX 5070.
Qwen3.5 9B Instruct (Q8)
Local OKOKB · 1
Best for: Quality, Coding, Reasoning·Pop: 86/100
Perf: est. 66 tok/s, typical error 5% · first token ~1.6s
Fits in 12 GB VRAM with room to spare. Best for quality, coding, reasoning on RTX 5070.
Qwen3.5 4B Instruct (Q8)
Local OKExcellentA · 60
Best for: Coding, Agents, Multimodal·Pop: 88/100
Perf: est. 113 tok/s, typical error 5% · first token ~0.8s
Fits in 12 GB VRAM with room to spare. Best for coding, agents, multimodal on RTX 5070.
Similar GPUs for local AI
Speeds are estimates for Qwen3 8B Q4_K_M, typical error 5%.
Compatible model families
- QwenAlibaba Cloud: Qwen3.8 27B local flagship (Apache 2.0, vision)
- LlamaMeta: Most popular open-weight model family
- DeepSeekDeepSeek AI: V4.1 Flash: multimodal 552B MoE, API only
- MistralMistral AI: Excellent performance-per-parameter ratio
- GemmaGoogle DeepMind: Gemma 4 E4B: 4.5B effective, runs on 8GB Macs
- PhiMicrosoft: Best quality-per-gigabyte at small sizes
RTX 5070 FAQ: common questions
How much VRAM does the RTX 5070 have for LLMs?
The RTX 5070 has 12GB GDDR7 VRAM with 672 GB/s bandwidth, 33% faster than the RTX 4070 SUPER. About 11.5GB is usable for model loading. Best suited for 7B-9B models at Q4 quantization.
What size LLM can I run on an RTX 5070?
Up to 9B parameter models at Q4 quantization. This includes Qwen 2.5 7B, Llama 3.2 8B, Mistral 7B, and Gemma 2 9B. The RTX 5070 runs 8B models at est. 98 tok/s, typical error 5%, the fastest 12GB card available.
Is the RTX 5070 good for local AI in 2026?
The RTX 5070 is the best 12GB card for local AI in 2026. Blackwell architecture and GDDR7 deliver est. 98 tok/s, typical error 5%, 96% faster than the RTX 3060. It is the value pick among 12GB cards, though the 2026 memory shortage has pushed every tier well above MSRP, so check the live price before committing.
Should I get RTX 5070 (12GB) or RTX 5070 Ti (16GB)?
If you want to run 14B models, get the 5070 Ti (16GB). If 7B-9B models are enough, the RTX 5070 costs noticeably less and still delivers 78% of the Ti speed. The Ti also has 33% more bandwidth (896 vs 672 GB/s).
RTX 5070 vs RTX 3060 for running AI models?
The RTX 5070 is 96% faster (est. 98 tok/s, typical error 5% vs est. 50 tok/s) with the same 12GB VRAM. However, the 3060 costs less, especially used. For budget builds, the 3060 remains excellent. For best speed at 12GB, the 5070 wins.
How fast is a 27B-class model on the RTX 5070?
The RTX 5070's 12GB of VRAM cannot fit a 32B model comfortably. The largest size class it fits is 14B, at est. 60 tok/s, typical error 15%. The largest individual model in the catalog that fits is Ministral 3 14B (14B).
Related guides & benchmarks
Sizing local AI? Start with RAM & VRAM
Browse all NVIDIA GPUs for AI
Real-world calibration
Have this hardware? Share a measured run.
Measure a fixed Ollama reference model on your machine. This separate, opt-in contribution does not change ModelFit's third-party SWE-Bench leaderboard; results appear on the measured-speed leaderboard after the next site update.
$ npx -y @wecko-ai/modelfit bench --submit --cleanup
Want personalized recommendations?
Use our interactive wizard to compare models across Apple Silicon and NVIDIA GPUs.