llm-bench.io Quality results on AMD GPUs

412 results · 6 sources · 412 measurements

Scores from different versions, task subsets, harnesses and scoring rules are not directly comparable. How results are reviewed

Documented results

24 selected tests


View Reportllm-bench.io	Mellum2.1-12B-A2.5B-ThinkingQ4_K_MWeights ↗	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	76.04 / 100LLM judge · Mean	1,622.33 tok/sPrefill · mean · per request	232.1 tok/sGeneration · mean · per request	GPU count unknownUnknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Ubuntu 26.04.1 LTS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.96+102 Prompt / output ? / ? tokens Context 131072 Concurrency 1 Date 2026-10-17 Full record →
View Reportllm-bench.io	Darwin-28B-REASON.i1IQ3_SRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	71 / 100LLM judge · Mean	256.2 tok/sPrefill · mean · per request	21.11 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS CachyOS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.87+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-06 Full record →
View Reportllm-bench.io	Qwen3.8-27BIQ4_XSWeights ↗Conflicting evidence	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	87.88 / 100LLM judge · Mean	547.47 tok/sPrefill · mean · per request	64.03 tok/sGeneration · mean · per request	GPU count unknownUnknown · 24 GB VRAM/GPU Configuration Memory 24 GB VRAM/GPU OS CachyOS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.80+98 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-09-27 Full record →
View Reportllm-bench.io	gmcoderQ8_0Repository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	76.08 / 100LLM judge · Mean	374.16 tok/sPrefill · mean · per request	41.69 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Professionnel 25H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.87+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-04 Full record →
View Reportllm-bench.io	Qwen3.5-9BQ4_K_MModel ref. ↗	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	62.5 / 100LLM judge · Mean	1,029.22 tok/sPrefill · mean · per request	70.18 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Pro 26H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.96+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-17 Full record →
View Reportllm-bench.io	Qwen3.8-27BQ4_K_MModel ref. ↗	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	81.27 / 100LLM judge · Mean	171.11 tok/sPrefill · mean · per request	5.78 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Pro 26H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.93+102 Prompt / output ? / ? tokens Context 16384 Concurrency 1 Date 2026-10-14 Full record →
View Reportllm-bench.io	Mellum2.1-12B-A2.5B-ThinkingQ8_0Weights ↗Conflicting evidence	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	68.61 / 100LLM judge · Mean	1,627.49 tok/sPrefill · mean · per request	179.54 tok/sGeneration · mean · per request	GPU count unknownUnknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Ubuntu 26.04.1 LTS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.96+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-17 Full record →
View Reportllm-bench.io	Darwin-28B-REASON.i1IQ3_XXSRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	73.1 / 100LLM judge · Mean	357.17 tok/sPrefill · mean · per request	38.41 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS CachyOS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.87+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-06 Full record →
View Reportllm-bench.io	cyber-tiel-coder-35b-a3b-mtpUD-IQ3_XXSRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	85.76 / 100LLM judge · Mean	1,258.3 tok/sPrefill · mean · per request	113.78 tok/sGeneration · mean · per request	GPU count unknownUnknown · 24 GB VRAM/GPU Configuration Memory 24 GB VRAM/GPU OS Windows 11 Pro 26H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.85+101 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-02 Full record →
View Reportllm-bench.io	ornith-1.5-9b@q6_kQ6_KRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	73.71 / 100LLM judge · Mean	732.23 tok/sPrefill · mean · per request	39.8 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Professionnel 25H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.59+97 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-09-15 Full record →
View Reportllm-bench.io	qwen3.8-27b-gsq-rco@iq2_xsIQ2_XSRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	87.25 / 100LLM judge · Mean	363.3 tok/sPrefill · mean · per request	27.97 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.89+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-08 Full record →
View Reportllm-bench.io	Mellum2.1-12B-A2.5B-ThinkingQ8_0Weights ↗	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	60.35 / 100LLM judge · Mean	1,513.7 tok/sPrefill · mean · per request	178.52 tok/sGeneration · mean · per request	GPU count unknownUnknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Ubuntu 26.04.1 LTS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.96+102 Prompt / output ? / ? tokens Context 131072 Concurrency 1 Date 2026-10-17 Full record →
View Reportllm-bench.io	Darwin-28B-REASON.i1IQ3_XXSRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	71.19 / 100LLM judge · Mean	816.67 tok/sPrefill · mean · per request	38.42 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS CachyOS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.87+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-06 Full record →
View Reportllm-bench.io	devstral-small-2:latestQ4_K_MRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	65.83 / 100LLM judge · Mean	—	10.41 tok/sGeneration · mean · per request	GPU count unknownUnknown · 24 GB VRAM/GPU Configuration Memory 24 GB VRAM/GPU OS Windows Engine Ollama Backend Not reported Harness llm-benchmark 0.3.0+52 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-04-13 Full record →
View Reportllm-bench.io	ornith-1.5-9b@q6_kQ6_KRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	63.25 / 100LLM judge · Mean	756.41 tok/sPrefill · mean · per request	39.11 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Professionnel 25H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.59+97 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-09-15 Full record →
View Reportllm-bench.io	qwen3.8-27b-gsq-rco@iq2_xsIQ2_XSRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	73.43 / 100LLM judge · Mean	448.83 tok/sPrefill · mean · per request	33.89 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.89+102 Prompt / output ? / ? tokens Context 32768 Concurrency 1 Date 2026-10-08 Full record →
View Reportllm-bench.io	gemma-4-26B-A4B-itUD-Q5_K_XLWeights ↗Conflicting evidence	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	75.76 / 100LLM judge · Mean	790.94 tok/sPrefill · mean · per request	69.55 tok/sGeneration · mean · per request	GPU count unknownUnknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Ubuntu 26.04.1 LTS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.92+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-13 Full record →
View Reportllm-bench.io	gemma4:12b-it-qatQ4_0Repository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	69.79 / 100LLM judge · Mean	1,153.42 tok/sPrefill · mean · per request	53.97 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Pro 25H2 Engine Ollama Backend Not reported Harness llm-benchmark 0.4.59+97 Prompt / output ? / ? tokens Context 8192 Concurrency 1 Date 2026-09-18 Full record →
View Reportllm-bench.io	devstral-small-2:latestQ4_K_MRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	62.5 / 100LLM judge · Mean	—	11.81 tok/sGeneration · mean · per request	GPU count unknownUnknown · 24 GB VRAM/GPU Configuration Memory 24 GB VRAM/GPU OS Windows Engine Ollama Backend Not reported Harness llm-benchmark 0.4.20+97 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-04-25 Full record →
View Reportllm-bench.io	qwen3.5-9b-mtp@q4_k_xl (checkpoint unspecified)Q4_K_XLModel ref. ↗	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	68.8 / 100LLM judge · Mean	219.05 tok/sPrefill · mean · per request	47.69 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Professionnel 25H2 Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.59+97 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-09-16 Full record →
View Reportllm-bench.io	qwen3.8-27b-gsq-rco@iq2_xsIQ2_XSRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	66.19 / 100LLM judge · Mean	450.9 tok/sPrefill · mean · per request	31.87 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows Engine LM Studio Backend Not reported Harness llm-benchmark 0.4.89+102 Prompt / output ? / ? tokens Context 65536 Concurrency 1 Date 2026-10-08 Full record →
View Reportllm-bench.io	gemma-4-26B-A4B-itUD-Q5_K_XLWeights ↗Conflicting evidence	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	74.23 / 100LLM judge · Mean	1,183.08 tok/sPrefill · mean · per request	69.78 tok/sGeneration · mean · per request	GPU count unknownUnknown · 32 GB VRAM/GPU Configuration Memory 32 GB VRAM/GPU OS Ubuntu 26.04.1 LTS Engine llama.cpp Backend Not reported Harness llm-benchmark 0.4.92+102 Prompt / output ? / ? tokens Context 131072 Concurrency 1 Date 2026-10-13 Full record →
View Reportllm-bench.io	gemma4:12b-it-qatQ4_0Repository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	68.4 / 100LLM judge · Mean	1,174.48 tok/sPrefill · mean · per request	54.47 tok/sGeneration · mean · per request	GPU count unknownUnknown · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Pro 25H2 Engine Ollama Backend Not reported Harness llm-benchmark 0.4.71+98 Prompt / output ? / ? tokens Context 8192 Concurrency 1 Date 2026-09-21 Full record →
View Reportllm-bench.io	gemma4:12bQ4_K_MRepository unresolved	llm-bench.io QualityFour sequential scenarios: agent workflow, code generation, role play, research	72.44 / 100LLM judge · Mean	—	63.05 tok/sGeneration · mean · per request	GPU count unknownUnknown · 24 GB VRAM/GPU Configuration Memory 24 GB VRAM/GPU OS Windows 11 IoT Enterprise 24H2 Engine Ollama Backend Not reported Harness llm-benchmark 0.4.36+97 Prompt / output ? / ? tokens Context 196608 Concurrency 1 Date 2026-08-16 Full record →

Versions and harnesses

Versions: Not reported
Harnesses: llm-benchmark 0.3.0+52, llm-benchmark 0.4.7+85, llm-benchmark 0.4.8+85, llm-benchmark 0.4.9+85, llm-benchmark 0.4.12+88, llm-benchmark 0.4.13+89, llm-benchmark 0.4.14+90, llm-benchmark 0.4.15+90, llm-benchmark 0.4.16+93, llm-benchmark 0.4.18+95, llm-benchmark 0.4.20+97, llm-benchmark 0.4.21+97, llm-benchmark 0.4.22+97, llm-benchmark 0.4.29+97, llm-benchmark 0.4.30+97, llm-benchmark 0.4.36+97, llm-benchmark 0.4.43+97, llm-benchmark 0.4.44+97, llm-benchmark 0.4.46+97, llm-benchmark 0.4.55+97, llm-benchmark 0.4.59+97, llm-benchmark 0.4.71+98, llm-benchmark 0.4.80+98, llm-benchmark 0.4.82+98, llm-benchmark 0.4.85+101, llm-benchmark 0.4.87+102, llm-benchmark 0.4.89+102, llm-benchmark 0.4.92+102, llm-benchmark 0.4.93+102, llm-benchmark 0.4.96+102, llm-benchmark 0.4.97+102, llm-benchmark 0.4.98+102

Explore all 412 results →

This is a selection across sources, not a ranking or a matched experiment. Reading benchmark reports

GPU benchmark pages: R9700 · RX 9060 XT · RX 7900 XTX · RX 7900 GRE · RX 7800 XT · RX 7600 XT