{"name":"Ministral-3 8B","arena":{"id":"ministral-3-8b-instruct-bf16","name":"Ministral-3 8B","vram":18,"params":8,"vendor":"mistral","quality":{"coding":{"bench":"LiveCodeBench","value":61.6},"science":{"bench":"GPQA-Diamond","value":66.8},"knowledge":{"bench":"MMLU","value":76.1}},"contextK":256},"notes":"Model-ID: mistralai/Ministral-3-8B-Instruct-2512. Container: vllm/vllm-openai:v0.26.0-aarch64-cu129-ubuntu2404. llama-benchy 0.4.0. Bronrun: 2026-08-07T02:59:50+02:00. KV-cache fp8; max_model_len 131072; gpu_memory_utilization 0,9. Prefix caching uit; async scheduling aan. Extra serverflags: --tokenizer_mode mistral --config_format mistral --load_format mistral --max-num-batched-tokens 8192. 10 van de 11 tests hebben een meting die volgens de huidige Arena-regels meetelt. Bij deze historische run is geen sanity-check vastgelegd. Dat bewijs ontbreekt, ook wanneer de snelheidsmeting volgens de huidige regels meetelt. Afgebroken: 11-rate-sweep.","order":14,"engine":"vLLM v0.26.0","verdict":"Gemeten op 2026-08-07 met vLLM v0.26.0. In de chattest met tien gelijktijdige verzoeken haalt dit profiel 24 tokens/s per gebruiker, met gemiddeld 1,25 seconden tot het eerste token. Bij 25k context stijgt die gemiddelde wachttijd naar 23,88 seconden. Bij deze historische run is geen sanity-check vastgelegd. Dat bewijs ontbreekt, ook wanneer de snelheidsmeting volgens de huidige regels meetelt. De capaciteitsmeting is afgebroken; daarmee is geen volledig capaciteitsplafond vastgesteld.","findings":{"broke":{"body":"Bij 25000 inputtokens en 256 outputtokens met tien gelijktijdige verzoeken is de gemiddelde wachttijd voor het eerste token 23,88 seconden. Daarna genereert het model gemiddeld 6,23 tokens/s per gebruiker. De meting alleen toont niet of geheugen, promptverwerking of scheduling de oorzaak van vertraging is.","title":"Wachttijd bij 25k context"},"costly":{"body":"De run verwerkt 0,41 verzoeken/s bij een ingestelde aankomstsnelheid van 1,5 verzoeken/s. 300 van de 300 verzoeken slagen. De p95-wachttijd voor het eerste token is 2,45 seconden.","title":"Verzoeken en wachttijd bij piekbelasting"},"worked":{"body":"Bij 1024 input- en 1024 outputtokens met tien gelijktijdige verzoeken meet ik gemiddeld 24 tokens/s per gebruiker. De gemiddelde wachttijd voor het eerste token is 1,25 seconden.","title":"Chat bij tien gelijktijdige verzoeken"},"surprised":{"body":"10 van de 11 tests hebben een meting die volgens de huidige Arena-regels meetelt. Bij deze historische run is geen sanity-check vastgelegd. Dat bewijs ontbreekt, ook wanneer de snelheidsmeting volgens de huidige regels meetelt. Afgebroken: 11-rate-sweep.","title":"Welke metingen meetellen"}},"hardware":"DGX Spark, NVIDIA GB10, 128 GB unified memory","modelUrl":"https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512","provider":"Mistral AI","precision":"BF16","parameters":"8B","provenance":{"modelId":"mistralai/Ministral-3-8B-Instruct-2512","runPath":"results/ministral-3/ministral-3-8b-instruct/bf16","validity":{"tests":{"01-chat":"complete","02-rag-8k":"complete","08-sharegpt":"complete","09-reasoning":"complete","04-multi-turn":"complete","11-rate-sweep":"aborted","03-long-output":"complete","05-big-context":"complete","10-monday-peak":"complete","07-office-baseline":"complete","06-concurrency-stress":"complete"},"sanity":"not_recorded"},"sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","servedName":"ministral-3-8b-instruct-bf16","generatedAt":"2026-08-07T02:59:50+02:00","runRevision":"a948cec23b054c62fef595bad3cdb284066922ac","vllmVersion":"0.26.0","suiteVersion":"2026-08","repositoryUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks","contractVersion":1,"contractRevision":"a948cec23b054c62fef595bad3cdb284066922ac","commandProvenance":"reconstructed_from_raw_artifacts","llamaBenchyVersion":"0.4.0"},"architecture":"Dense","measuredContextK":128,"benchmarkResults":[{"id":"ministral-3-8b-instruct-bf16-chat-1k","date":"2026-08-07","runs":3,"metrics":{"ttfrSec":{"mean":1.248,"stddev":0.286},"decodeTokensPerSecondTotal":{"mean":183.89,"stddev":2.62},"prefillTokensPerSecondTotal":{"mean":6678.61,"stddev":168.8},"decodeTokensPerSecondPerUser":{"mean":24,"stddev":0.29}},"sourceRun":"Test 01","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","workloadId":"chat-1k","sourceLabel":"mistralai/Ministral-3-8B-Instruct-2512 bf16 op de DGX Spark","sourceStatus":"complete","sourceTestId":"01-chat","prefixCaching":false},{"id":"ministral-3-8b-instruct-bf16-rag-8k","date":"2026-08-07","runs":3,"metrics":{"ttfrSec":{"mean":6.335,"stddev":3.128},"decodeTokensPerSecondTotal":{"mean":124.14,"stddev":1.82},"prefillTokensPerSecondTotal":{"mean":6877.56,"stddev":21.23},"decodeTokensPerSecondPerUser":{"mean":14.74,"stddev":1.42}},"sourceRun":"Test 02","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","workloadId":"rag-8k","sourceLabel":"mistralai/Ministral-3-8B-Instruct-2512 bf16 op de DGX Spark","sourceStatus":"complete","sourceTestId":"02-rag-8k","prefixCaching":false},{"id":"ministral-3-8b-instruct-bf16-long-output-agents","date":"2026-08-07","runs":3,"metrics":{"ttfrSec":{"mean":0.647,"stddev":0.005},"decodeTokensPerSecondTotal":{"mean":117.83,"stddev":7.95},"prefillTokensPerSecondTotal":{"mean":3611.89,"stddev":36.89},"decodeTokensPerSecondPerUser":{"mean":25.27,"stddev":0.15}},"sourceRun":"Test 03","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","workloadId":"long-output-agents","sourceLabel":"mistralai/Ministral-3-8B-Instruct-2512 bf16 op de DGX Spark","sourceStatus":"complete","sourceTestId":"03-long-output","prefixCaching":false},{"id":"ministral-3-8b-instruct-bf16-multi-turn-office","date":"2026-08-07","runs":3,"metrics":{"ttfrSec":{"mean":1.966,"stddev":0.701},"decodeTokensPerSecondTotal":{"mean":206.49,"stddev":0.76},"prefillTokensPerSecondTotal":{"mean":7081.71,"stddev":533.5},"decodeTokensPerSecondPerUser":{"mean":21.89,"stddev":0.6}},"sourceRun":"Test 04","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","workloadId":"multi-turn-office","sourceLabel":"mistralai/Ministral-3-8B-Instruct-2512 bf16 op de DGX Spark","sourceStatus":"complete","sourceTestId":"04-multi-turn","prefixCaching":false},{"id":"ministral-3-8b-instruct-bf16-large-context-25k","date":"2026-08-07","runs":3,"metrics":{"ttfrSec":{"mean":23.875,"stddev":12.352},"decodeTokensPerSecondTotal":{"mean":39.53,"stddev":0.15},"prefillTokensPerSecondTotal":{"mean":5237.48,"stddev":14.47},"decodeTokensPerSecondPerUser":{"mean":6.23,"stddev":1.83}},"sourceRun":"Test 05","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","workloadId":"large-context-25k","sourceLabel":"mistralai/Ministral-3-8B-Instruct-2512 bf16 op de DGX Spark","sourceStatus":"complete","sourceTestId":"05-big-context","prefixCaching":false},{"id":"ministral-3-8b-instruct-bf16-concurrency-stress-25k","date":"2026-08-07","runs":3,"metrics":{"ttfrSec":{"mean":45.792,"stddev":25.329},"decodeTokensPerSecondTotal":{"mean":40.99,"stddev":0.09},"prefillTokensPerSecondTotal":{"mean":5127.4,"stddev":8.14},"decodeTokensPerSecondPerUser":{"mean":3.48,"stddev":1.17}},"sourceRun":"Test 06","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","workloadId":"concurrency-stress-25k","sourceLabel":"mistralai/Ministral-3-8B-Instruct-2512 bf16 op de DGX Spark","sourceStatus":"complete","sourceTestId":"06-concurrency-stress","prefixCaching":false}],"openLoopResults":[{"id":"ministral-3-8b-instruct-bf16-office-random-4k","date":"2026-08-07","metrics":{"e2eMs":{"p50":33401.72,"p90":58981.42,"p95":66852.52,"p99":70552.98,"mean":34227.01},"tpotMs":{"p50":68.77,"p90":82.56,"p95":87.64,"p99":98.63,"mean":68.13},"ttftMs":{"p50":1286.9,"p90":2118.63,"p95":2386.53,"p99":3125.93,"mean":1347.19}},"sourceRun":"Test 07","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","scenarioId":"office-random-4k","achievedRps":0.286,"sourceStatus":"complete","sourceTestId":"07-office-baseline","configuredRps":0.3,"prefixCaching":false,"totalRequests":200,"successfulRequests":200,"totalTokenThroughput":1299.8,"peakConcurrentRequests":18},{"id":"ministral-3-8b-instruct-bf16-sharegpt-replay","date":"2026-08-07","metrics":{"e2eMs":{"p50":5640.94,"p90":19104.78,"p95":25193.2,"p99":31958.9,"mean":8348.17},"tpotMs":{"p50":38.89,"p90":40.85,"p95":42.04,"p99":45.58,"mean":39.26},"ttftMs":{"p50":136.29,"p90":461.11,"p95":509.26,"p99":607.12,"mean":193.43}},"sourceRun":"Test 08","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","scenarioId":"sharegpt-replay","achievedRps":0.292,"sourceStatus":"complete","sourceTestId":"08-sharegpt","configuredRps":0.3,"prefixCaching":false,"totalRequests":250,"successfulRequests":250,"totalTokenThroughput":125.04,"peakConcurrentRequests":10},{"id":"ministral-3-8b-instruct-bf16-reasoning","date":"2026-08-07","metrics":{"e2eMs":{"p50":277532.72,"p90":465512.62,"p95":498529.79,"p99":519225.63,"mean":291868.53},"tpotMs":{"p50":70.66,"p90":77.1,"p95":78.18,"p99":78.99,"mean":68.84},"ttftMs":{"p50":591.8,"p90":715.43,"p95":751.83,"p99":997.7,"mean":528.84}},"sourceRun":"Test 09","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","scenarioId":"reasoning","achievedRps":0.072,"sourceStatus":"complete","sourceTestId":"09-reasoning","configuredRps":0.2,"prefixCaching":false,"totalRequests":50,"successfulRequests":50,"totalTokenThroughput":379.65,"peakConcurrentRequests":43},{"id":"ministral-3-8b-instruct-bf16-monday-peak-random-4k","date":"2026-08-07","metrics":{"e2eMs":{"p50":56937.13,"p90":103310.29,"p95":106877.99,"p99":113469.58,"mean":58079.29},"tpotMs":{"p50":115.8,"p90":127.64,"p95":131.51,"p99":150.61,"mean":114.52},"ttftMs":{"p50":1219.37,"p90":1974.09,"p95":2445.41,"p99":4338.82,"mean":1334.4}},"sourceRun":"Test 10","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","scenarioId":"monday-peak-random-4k","achievedRps":0.414,"sourceStatus":"complete","sourceTestId":"10-monday-peak","configuredRps":1.5,"prefixCaching":false,"totalRequests":300,"successfulRequests":300,"totalTokenThroughput":1862.01,"peakConcurrentRequests":27}],"rateSweep":{"date":"2026-08-07","capacity":{"slo2000":null,"slo5000":{"outputTps":47.4,"ttftP95Ms":2674.8,"achievedRps":0.096,"configuredRps":0.1,"peakConcurrent":8},"slo10000":{"outputTps":47.4,"ttftP95Ms":2674.8,"achievedRps":0.096,"configuredRps":0.1,"peakConcurrent":8}},"sourceRun":"Test 11","sourceUrl":"https://github.com/djangodevreng/dgx-spark-benchmarks/tree/a948cec23b054c62fef595bad3cdb284066922ac/results/ministral-3/ministral-3-8b-instruct/bf16/","abortedAtRps":0.2,"queueKneeRps":null,"sourceStatus":"aborted","sourceTestId":"11-rate-sweep","totalRequests":100,"successfulRequests":100}}