Benchmark result

qwen3.8:latest on AMD Radeon RX 7900 XTX — 41.5 tok/s

Measured with Ollama 0.32.12 on August 14, 2026.

ShareRedditX

LLM / Model

Model Size
27.3B
Architecture
qwen
Quantization
Q4_K_M
Memory Kind
model
Tool
Ollama v0.32.12
Scenarios

LLM Quality Assessment

Agent WorkflowopenclawCOMPLETED
Overall Quality Score
80.40
coherence
82
decomposition
85
error_handling
74
tool_selection
78

The plan is logically structured and covers the required report components, with a clear progression from data gathering to synthesis and delivery. Tool choices are mostly appropriate, though send_email is unnecessary for producing the report and the execute_code step is doing more than analysis would require. Decomposition is strong, with the task split into manageable stages and explicit dependencies. Error handling is decent due to a stated backup strategy, but the fallback approach is somewhat vague and depends on hard-coded constants rather than a precise retrieval contingency.

Code Generationcoding_agentCOMPLETED▶ Play Game
Overall Quality Score
80.60
correctness
82
performance
76
code_quality
84
completeness
74

A solid single-file Breakout implementation with good visuals and basic mobile/keyboard support, but it misses some strict layout and collision-spec requirements, especially brick placement, exact AABB side resolution, and a proper serve/start state.

Role Play & NarrativeroleplayCOMPLETED
Overall Quality Score
84.25
dialogue
84
immersion
95
consistency
88
narrative_arc
52

Highly atmospheric and emotionally credible as a single scene, with excellent sensory detail and a convincing reaction to Kestra, but it falls short of the required multi-turn arc and final help/refuse commitment.

Research & AnalysisresearchCOMPLETED
Overall Quality Score
86.90
depth
89
clarity
91
insight
86
interpretation
84

Strong, well-structured analysis with good task-by-task comparisons, justified recommendations, and explicit caveats. It is especially clear and actionable, but some interpretation slightly overstates pure parameter-scaling effects because architecture and evaluation conditions are confounded.

GPU Acceleration

Processor Offload
100% GPU
GPU Model
AMD Radeon RX 7900 XTX
VRAM
23 GB

Performance Metrics

Generation Speed
41.5
tokens/sec
TTFT
22796
milliseconds
Memory Usage
16.33
GB
Memory (Raw)
16724
MB

Context Information

Tokens Used
6550
Context Window
65,536

CPU & Memory

CPU Model
AMD Ryzen 7 7800X3D 8-Core Processor
CPU Threads
16
Physical Cores
8
CPU Frequency
N/A
Total RAM
31 GB

Operating System

OS Name
windows
OS Version
Windows 11 Home 25H2
Kernel Version
Microsoft Windows [Version 10.0.26200.9168]

Submission Details

Submitted:
8/14/2026, 6:22:05 PM
Client:
v0.4.36+97
Benchmark Result ID:
cmst9xr50000f01p4l8j7imx1