Skip to content
m.recipes
Find a configuration
Sign in
96 GB · 28 reports

RTX PRO 6000 Blackwell Workstation Edition · 96 GB

Qualified results first; original reports and full tables stay below.
Reset

Text generation · Qualified results

Speeds compare only within one exact test. Recipe availability is checked on each report against the published recipe executor.

No qualified results yet · Text generation · RTX PRO 6000 Blackwell Workstation Edition · 96 GB

28 original reports are kept as evidence but do not qualify for comparison; each report states why. This does not show that the model or GPU cannot run it.

Check model launch supportView original reportsMeasured alternatives for this taskTest your own GPU

Measured alternatives on other GPUs

Measured on different GPUs with the same or smaller listed capacity. These are references for choosing hardware, not performance or compatibility evidence for your selected GPU.

  1. WeVisOne-VL-2B

    tencent · 6 tested configurations · 1 GPU

    • RTX PRO 6000 Blackwell Server Edition · 96 GBBF16 · 32K → 2K tokens · No cache reuse
      Output throughput · includes prefill · 1 request
      153tok/s
      Best measured · 16 concurrent requests
      319tok/s
      Observed GPU occupancy
      85.3GiB
      First token
      753 ms
      Use this recipe · 16 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
      Test details · 6 concurrency levels

      Total output tokens divided by the full request window, including prefill and queueing.

      Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

      Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

      Engine
      vllm · vLLM 0.28.0
      Protocol
      wbench-search-http-v1
      Dataset
      migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
      Workload
      32768in-2048out
      ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
      C1153.4413.39 s753 ms85.25343.41d2914dd8 ↗
      C2208.6519.64 s1.1 s85.25360.41d2914dd8 ↗
      C4259.1933.21 s1.6 s85.25398.91d2914dd8 ↗
      C8295.3759.5 s2.1 s85.254221d2914dd8 ↗
      C12311.3386.14 s2.7 s85.25432.41d2914dd8 ↗
      C16319.17112.4 s3.3 s85.254381d2914dd8 ↗
    • 5 other test configurations on this GPU
      • BF16 · 32K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        162tok/s
        Best measured · 18 concurrent requests
        857tok/s
        Observed GPU occupancy
        85.3GiB
        First token
        71 ms
        Use this recipe · 18 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 7 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        32768in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C1161.5712.72 s71 ms85.25327.8175198c63 ↗
        C2269.4315.21 s86 ms85.25349.1175198c63 ↗
        C4460.4117.83 s106 ms85.25427.3175198c63 ↗
        C8644.4525.52 s127 ms85.25486.9175198c63 ↗
        C16771.0442.77 s184 ms85.25533.9175198c63 ↗
        C17812.7943.14 s195 ms85.25550.4175198c63 ↗
        C18856.6443.41 s196 ms85.25568.9175198c63 ↗
      • BF16 · 64K → 2K tokens · No cache reuse
        Output throughput · includes prefill · 1 request
        101tok/s
        Best measured · 8 concurrent requests
        151tok/s
        Observed GPU occupancy
        85.3GiB
        First token
        2.3 s
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 5 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
        Workload
        65536in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C1101.2620.26 s2.3 s85.25369.11c5c9cf6f ↗
        C2123.1133.41 s3.6 s85.25404.81c5c9cf6f ↗
        C4140.1964.26 s4.5 s85.254301c5c9cf6f ↗
        C6142.0897.79 s5.3 s85.25430.41c5c9cf6f ↗
        C8150.73122.37 s6.3 s85.25446.71c5c9cf6f ↗
      • BF16 · 64K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        114tok/s
        Best measured · 9 concurrent requests
        433tok/s
        Observed GPU occupancy
        85.3GiB
        First token
        135 ms
        Use this recipe · 9 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 7 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        65536in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C1113.5118.06 s135 ms85.25341.315fee1315 ↗
        C2153.326.73 s161 ms85.25369.515fee1315 ↗
        C4281.8629.1 s202 ms85.25483.815fee1315 ↗
        C6293.7441.93 s214 ms85.25459.215fee1315 ↗
        C7340.142.27 s227 ms85.25496.115fee1315 ↗
        C8387.9342.41 s237 ms85.25534.915fee1315 ↗
        C9433.0742.78 s246 ms85.25575.815fee1315 ↗
      • BF16 · 128K → 2K tokens · No cache reuse
        Output throughput · includes prefill · 1 request
        56.1tok/s
        Best measured · 4 concurrent requests
        66.8tok/s
        Observed GPU occupancy
        85.3GiB
        First token
        8 s
        Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 4 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
        Workload
        131072in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C156.0936.63 s8 s85.25418.912c661c7e ↗
        C262.5267.49 s12 s85.25444.312c661c7e ↗
        C363.59109.31 s13.5 s85.25446.812c661c7e ↗
        C466.75144.63 s15 s85.25461.612c661c7e ↗
      • BF16 · 128K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        71tok/s
        Best measured · 4 concurrent requests
        165tok/s
        Observed GPU occupancy
        85.3GiB
        First token
        268 ms
        Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 4 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        131072in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C171.0528.83 s268 ms85.25356.611ff0c98a ↗
        C289.4345.82 s323 ms85.25399.611ff0c98a ↗
        C3135.4345.41 s379 ms85.25470.911ff0c98a ↗
        C4165.2949.61 s420 ms85.2552411ff0c98a ↗

  2. Qwen3.8-27B-NVFP4

    unsloth · 11 tested configurations · 1 GPU

    • RTX PRO 6000 Blackwell Server Edition · 96 GBNVFP4 · 32K → 2K tokens · No cache reuse
      Output throughput · includes prefill · 1 request
      52.8tok/s
      Best measured · 8 concurrent requests
      215tok/s
      Observed GPU occupancy
      87.4GiB
      First token
      3.3 s
      Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
      Test details · 6 concurrency levels

      Total output tokens divided by the full request window, including prefill and queueing.

      Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

      Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

      Engine
      vllm · vLLM 0.28.0
      Protocol
      wbench-search-http-v1
      Dataset
      migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
      Workload
      32768in-2048out
      ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
      C152.7838.8 s3.3 s87.37337.31aee8b33e ↗
      C287.248.71 s4.9 s87.37343.41aee8b33e ↗
      C4142.1464.91 s7 s87.37375.31aee8b33e ↗
      C6187.6979.85 s8.1 s87.37402.81aee8b33e ↗
      C7202.3185.21 s8.7 s87.37411.81aee8b33e ↗
      C8215.0594.52 s9.3 s87.37421.41aee8b33e ↗
    • 10 other test configurations on this GPU
      • NVFP4 · 32K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        57tok/s
        Best measured · 8 concurrent requests
        374tok/s
        Observed GPU occupancy
        87.4GiB
        First token
        222 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 6 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        32768in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C156.9636.17 s222 ms87.37322.61bced0642 ↗
        C2104.1439.34 s355 ms87.37305.91bced0642 ↗
        C4198.4241.3 s704 ms87.37319.31bced0642 ↗
        C6291.1242.23 s1 s87.37342.11bced0642 ↗
        C7333.3243.04 s1.1 s87.37349.11bced0642 ↗
        C8374.0443.82 s1.3 s87.37353.21bced0642 ↗
      • NVFP4 · 32K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        57.1tok/s
        Best measured · 8 concurrent requests
        373tok/s
        Observed GPU occupancy
        86.6GiB
        First token
        231 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 3 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        synthetic-tokenizer-v3:cache=prefix-reuse:repeats=3
        Workload
        32768in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C157.0935.88 s231 ms86.63319137f29252 ↗
        C4197.841.44 s650 ms86.63330.2137f29252 ↗
        C8372.8843.98 s1.3 s86.63367.3137f29252 ↗
      • NVFP4 · 32K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        56.4tok/s
        Best measured · 8 concurrent requests
        371tok/s
        Observed GPU occupancy
        86.4GiB
        First token
        244 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 3 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        synthetic-tokenizer-v3:cache=prefix-reuse:repeats=3
        Workload
        32768in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C156.436.51 s244 ms86.35319.11791b887d ↗
        C4196.4541.73 s654 ms86.35315.51791b887d ↗
        C8370.6844.24 s1.3 s86.36348.11791b887d ↗
      • NVFP4 · 64K → 2K tokens · No cache reuse
        Output throughput · includes prefill · 1 request
        44.9tok/s
        Best measured · 4 concurrent requests
        96.8tok/s
        Observed GPU occupancy
        87.4GiB
        First token
        8.6 s
        Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 4 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
        Workload
        65536in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C144.8645.67 s8.6 s87.37360.51d2ff32ec ↗
        C267.5963.21 s12.7 s87.37379.11d2ff32ec ↗
        C384.9584.07 s14.6 s87.37398.51d2ff32ec ↗
        C496.77105.62 s16.3 s87.37415.81d2ff32ec ↗
      • NVFP4 · 64K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        54.7tok/s
        Best measured · 8 concurrent requests
        340tok/s
        Observed GPU occupancy
        87.4GiB
        First token
        329 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 6 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        65536in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C154.7237.56 s329 ms87.37322.41d3f6b770 ↗
        C299.9540.99 s522 ms87.37315.91d3f6b770 ↗
        C4189.0743.34 s971 ms87.37337.81d3f6b770 ↗
        C6264.746.44 s1.3 s87.373641d3f6b770 ↗
        C7303.0847.32 s1.4 s87.37373.41d3f6b770 ↗
        C8339.6148.27 s1.7 s87.37381.21d3f6b770 ↗
      • NVFP4 · 64K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        54tok/s
        Best measured · 8 concurrent requests
        337tok/s
        Observed GPU occupancy
        86.4GiB
        First token
        361 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 3 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        synthetic-tokenizer-v3:cache=prefix-reuse:repeats=3
        Workload
        65536in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C154.0438.07 s361 ms86.36320.71417f69be ↗
        C4187.1243.81 s879 ms86.36333.21417f69be ↗
        C8336.9548.66 s1.6 s86.36379.71417f69be ↗
      • NVFP4 · 64K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        54.5tok/s
        Best measured · 8 concurrent requests
        339tok/s
        Observed GPU occupancy
        86.6GiB
        First token
        359 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 3 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        synthetic-tokenizer-v3:cache=prefix-reuse:repeats=3
        Workload
        65536in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C154.5437.55 s359 ms86.63324.917d1a7c8e ↗
        C4188.4843.48 s878 ms86.63350.217d1a7c8e ↗
        C8338.6548.43 s1.6 s86.63396.917d1a7c8e ↗
      • NVFP4 · 128K → 2K tokens · No cache reuse
        Output throughput · includes prefill · 1 request
        31.5tok/s
        Best measured · 2 concurrent requests
        41.3tok/s
        Observed GPU occupancy
        87.4GiB
        First token
        24.9 s
        Use this recipe · 2 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 2 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-search-http-v1
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
        Workload
        131072in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C131.5464.93 s24.9 s87.37402.7173fcb7c1 ↗
        C241.27111.14 s34 s87.37426.3173fcb7c1 ↗
      • NVFP4 · 128K → 2K tokens · No cache reuse
        Output throughput · includes prefill · 1 request
        31.5tok/s
        Best measured · 8 concurrent requests
        57.7tok/s
        Observed GPU occupancy
        86.6GiB
        First token
        24.9 s
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 3 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        synthetic-tokenizer-v3:cache=isolated:repeats=3
        Workload
        131072in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C131.4565.12 s24.9 s86.63415.71448b0777 ↗
        C450.67220.6 s45.8 s86.63469.41448b0777 ↗
        C857.74421.65 s61.9 s86.63494.51448b0777 ↗
      • NVFP4 · 128K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        50tok/s
        Best measured · 8 concurrent requests
        287tok/s
        Observed GPU occupancy
        86.4GiB
        First token
        506 ms
        Use this recipe · 8 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 3 concurrency levels

        Total output tokens divided by the full request window, including prefill and queueing.

        Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

        Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.

        Engine
        vllm · vLLM 0.28.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        synthetic-tokenizer-v3:cache=prefix-reuse:repeats=3
        Workload
        131072in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C149.9840.98 s506 ms86.36322.7168d1f2c6 ↗
        C4171.3347.85 s1 s86.36365.5168d1f2c6 ↗
        C8287.3157.12 s1.9 s86.36425.4168d1f2c6 ↗

Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

Measurements without a comparable protocol · 242 rows

These measured rates remain visible, but are excluded from comparison curves because the protocol is unknown.

Measurements without a comparable protocol · 242 rows
ModelFormatEngineWorkloadConcurrencyThroughputGPU power (W)
Qwen3-0.6B-FP8 ↗FP8vLLM2K1322.44 tok/s210.89
Qwen3-0.6B-FP8 ↗FP8vLLM2K1322.88 tok/s212.86
Qwen3-0.6B-FP8 ↗FP8vLLM2K82,273.77 tok/s255.08
Qwen3-0.6B-FP8 ↗FP8vLLM2K82,273.93 tok/s255.49
Qwen3-0.6B-FP8 ↗FP8vLLM2K325,859.12 tok/s272.37
Qwen3-0.6B-FP8 ↗FP8vLLM2K325,843.57 tok/s273.35
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K1144.82 tok/s278.5
Qwen3-1.7B-FP8 ↗FP8vLLM2K1216.01 tok/s279.87
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K1148.58 tok/s280.97
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K1148.76 tok/s282.63
Qwen3-VL-2B-FP8 ↗FP8vLLM2K1210.45 tok/s286.51
Qwen3-0.6B-FP8 ↗FP8vLLM2K164,021.54 tok/s297.98
Qwen3-0.6B-FP8 ↗FP8vLLM2K164,032.86 tok/s300.74
Qwen3-1.7B-FP8 ↗FP8vLLM2K81,517.15 tok/s301.04
Qwen3-VL-2B-FP8 ↗FP8vLLM2K81,347.81 tok/s303.39
Qwen3-0.6B-FP8 ↗FP8vLLM2K649,169.25 tok/s310.65
Qwen3-0.6B-FP8 ↗FP8vLLM2K649,180.11 tok/s310.96
Qwen3-0.6B ↗NativevLLM2K1437.51 tok/s311.24
Qwen3-0.6B ↗NativevLLM2K82,567.27 tok/s313.85
Qwen3-VL-2B-FP8 ↗FP8vLLM2K323,919.61 tok/s319.5
Qwen3-1.7B-FP8 ↗FP8vLLM2K323,995.02 tok/s322.48
Qwen3-VL-2B-FP8 ↗FP8vLLM2K162,567.25 tok/s333.03
Qwen3-1.7B-FP8 ↗FP8vLLM2K162,767.17 tok/s334.69
Qwen3-VL-4B-FP8 ↗FP8vLLM2K1121.03 tok/s336.8
Qwen3-4B-FP8 ↗FP8vLLM2K1123.74 tok/s339.99
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K1123.78 tok/s340.7
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K1124.1 tok/s340.71
Qwen3-VL-2B-Instruct ↗NativevLLM2K81,505.94 tok/s343.21
Qwen3-VL-4B-FP8 ↗FP8vLLM2K8840.77 tok/s344.48
Qwen3-0.6B ↗NativevLLM2K325,923.76 tok/s344.82
Qwen3-VL-4B-FP8 ↗FP8vLLM2K322,440.48 tok/s346.24
Qwen3-1.7B ↗NativevLLM2K81,653.34 tok/s346.29
Qwen3-4B-FP8 ↗FP8vLLM2K8900.5 tok/s348.42
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K8825.93 tok/s349.18
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K8849.15 tok/s349.68
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K1148.1 tok/s350.08
Qwen3-VL-2B-FP8 ↗FP8vLLM2K646,041.38 tok/s350.25
Qwen3-0.6B ↗NativevLLM2K164,109.44 tok/s351.47
Qwen3-30B-A3B ↗NativevLLM2K1153.75 tok/s355.98
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K322,568.27 tok/s356.3
Qwen3-1.7B-FP8 ↗FP8vLLM2K646,638.25 tok/s357.34
Qwen3-4B-FP8 ↗FP8vLLM2K322,461.61 tok/s357.75
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K322,553.39 tok/s358.2
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K8633.5 tok/s362.33
Qwen3-VL-4B-FP8 ↗FP8vLLM2K161,474.4 tok/s363.33
Qwen3-VL-2B-Instruct ↗NativevLLM2K324,101.26 tok/s367.33
Qwen3-0.6B-FP8 ↗FP8vLLM2K12812,717.18 tok/s367.48
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K8569.2 tok/s367.88
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K161,605.45 tok/s373.79
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K8612.01 tok/s374.13
Qwen3-0.6B-FP8 ↗FP8vLLM2K12812,746.39 tok/s374.65
Qwen3-4B-FP8 ↗FP8vLLM2K161,689.18 tok/s374.68
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K161,526.88 tok/s374.72
Qwen3-VL-4B-FP8 ↗FP8vLLM2K644,134.1 tok/s375.46
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K8420.47 tok/s376.63
Qwen3-VL-4B-Instruct ↗NativevLLM2K8848.21 tok/s376.81
Qwen3-1.7B ↗NativevLLM2K324,469.59 tok/s380.55
Qwen3-VL-2B-Instruct ↗NativevLLM2K162,702.09 tok/s380.63
Qwen3-30B-A3B ↗NativevLLM2K8443.44 tok/s381.16
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K644,192.4 tok/s384.46
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K644,391.43 tok/s384.86
Qwen3-4B ↗NativevLLM2K8922.46 tok/s385.95
Qwen3-4B-FP8 ↗FP8vLLM2K644,589.22 tok/s386.08
Qwen3-1.7B ↗NativevLLM2K162,961.26 tok/s389.12
Qwen3-VL-8B-FP8 ↗FP8vLLM2K8532.73 tok/s392.3
Qwen3-VL-2B-FP8 ↗FP8vLLM2K1288,188.52 tok/s393.61
Qwen3-0.6B ↗NativevLLM2K648,276.23 tok/s395.15
Qwen3-VL-8B-FP8 ↗FP8vLLM2K184.34 tok/s396.19
Qwen3-1.7B-FP8 ↗FP8vLLM2K1288,870.96 tok/s396.88
Qwen3-VL-2B-Instruct ↗NativevLLM2K1262.38 tok/s398.62
Qwen3-8B-FP8 ↗FP8vLLM2K8630.94 tok/s399.14
Qwen3-VL-8B-FP8 ↗FP8vLLM2K321,929.49 tok/s400.07
Qwen3-8B-FP8 ↗FP8vLLM2K185.3 tok/s400.47
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K161,011.04 tok/s400.67
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K16586.67 tok/s405.52
Qwen3-30B-A3B ↗NativevLLM2K16670.83 tok/s407.18
Qwen3-1.7B ↗NativevLLM2K1269.85 tok/s407.98
Qwen3-8B-FP8 ↗FP8vLLM2K322,148.83 tok/s408.74
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K16899.68 tok/s409.66
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K16975.61 tok/s411.19
Qwen3-VL-4B-Instruct ↗NativevLLM2K161,567.84 tok/s412.02
Qwen3-VL-8B-FP8 ↗FP8vLLM2K161,033.23 tok/s414.66
Qwen3-VL-8B-FP8 ↗FP8vLLM2K643,186.24 tok/s417.02
Qwen3-4B-FP8 ↗FP8vLLM2K1285,914.47 tok/s420.27
Qwen3-VL-8B-Instruct ↗NativevLLM2K8532.81 tok/s420.32
Qwen3-8B-FP8 ↗FP8vLLM2K161,245.82 tok/s421.56
Qwen3-4B ↗NativevLLM2K161,689.63 tok/s423.75
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K1286,005.61 tok/s424.98
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K1286,015.29 tok/s425.21
Qwen3-8B-FP8 ↗FP8vLLM2K643,110.1 tok/s425.33
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K642,466.69 tok/s425.69
Qwen3-VL-2B-Instruct ↗NativevLLM2K646,187.38 tok/s425.87
Qwen3-0.6B-FP8 ↗FP8vLLM2K25615,433.69 tok/s427.13
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K642,504.94 tok/s427.44
Qwen3-8B ↗NativevLLM2K8611.18 tok/s428.44
Qwen3-0.6B ↗NativevLLM2K1289,915.75 tok/s428.69
Qwen3-VL-4B-FP8 ↗FP8vLLM2K1286,168.1 tok/s428.99
Qwen3-0.6B-FP8 ↗FP8vLLM2K25615,410.03 tok/s429.66
Qwen3-VL-4B-Instruct ↗NativevLLM2K322,573.74 tok/s432.46
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K642,174.46 tok/s434.22
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K321,014.41 tok/s437.08
Qwen3-1.7B-FP8 ↗FP8vLLM2K2569,976.22 tok/s438.57
Qwen3-4B ↗NativevLLM2K322,609.22 tok/s438.64
Qwen3-30B-A3B ↗NativevLLM2K321,077.51 tok/s438.97
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K321,677.68 tok/s441.34
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K641,471.14 tok/s441.59
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K321,410.74 tok/s442.67
Qwen3-1.7B ↗NativevLLM2K646,736.75 tok/s443.26
Qwen3-VL-2B-FP8 ↗FP8vLLM2K2569,682.31 tok/s443.89
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K321,585.35 tok/s448.08
Qwen3-30B-A3B ↗NativevLLM2K641,452.88 tok/s448.3
Qwen3-14B ↗NativevLLM2K8306.62 tok/s453.57
Qwen3-0.6B ↗NativevLLM2K25610,892.38 tok/s453.68
Qwen3-0.6B ↗NativevLLM2K102411,531.55 tok/s454.04
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K1284,010.41 tok/s455.17
Qwen3-VL-8B-Instruct ↗NativevLLM2K16972.99 tok/s455.98
Qwen3-8B-FP8 ↗FP8vLLM2K1285,213.44 tok/s457.23
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K1283,900.43 tok/s459.45
Qwen3-8B ↗NativevLLM2K161,193.17 tok/s461.06
Qwen3-VL-8B-Instruct ↗NativevLLM2K321,612.8 tok/s461.36
Qwen3-1.7B-FP8 ↗FP8vLLM2K102410,854.98 tok/s463.2
Qwen3-VL-8B-FP8 ↗FP8vLLM2K1284,866.51 tok/s464.46
Qwen3-8B ↗NativevLLM2K321,621.49 tok/s464.82
Qwen3-14B-FP8 ↗FP8vLLM2K8394.41 tok/s465.01
Qwen3-1.7B-FP8 ↗FP8vLLM2K51210,327.11 tok/s465.3
Qwen3-0.6B ↗NativevLLM2K51211,461.61 tok/s465.32
Qwen3-VL-2B-FP8 ↗FP8vLLM2K51210,320.27 tok/s465.7
Qwen3-VL-32B-Instruct ↗NativevLLM2K8160.3 tok/s467.36
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K1283,403.48 tok/s467.41
Qwen3-32B ↗NativevLLM2K8133.48 tok/s468.39
Qwen3-14B-FP8 ↗FP8vLLM2K321,344.94 tok/s469.25
Qwen3-14B-FP8 ↗FP8vLLM2K160.04 tok/s469.66
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K1282,176.68 tok/s469.99
Qwen3-VL-2B-FP8 ↗FP8vLLM2K102410,950.7 tok/s470
Qwen3-VL-2B-FP8 ↗FP8vLLM2K153610,600.83 tok/s470.88
Qwen3-0.6B-FP8 ↗FP8vLLM2K102416,940.56 tok/s471.41
Qwen3-0.6B-FP8 ↗FP8vLLM2K102416,919.17 tok/s472.38
Qwen3-VL-2B-Instruct ↗NativevLLM2K1287,744.92 tok/s475.19
Qwen3-0.6B-FP8 ↗FP8vLLM2K51217,178.11 tok/s475.21
Qwen3-VL-2B-Instruct ↗NativevLLM2K2569,490.99 tok/s475.4
Qwen3-0.6B-FP8 ↗FP8vLLM2K51217,177.4 tok/s475.67
Qwen3-0.6B-FP8 ↗FP8vLLM2K153617,189.94 tok/s475.84
Qwen3-1.7B ↗NativevLLM2K102410,387.25 tok/s477.36
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K2563,052.25 tok/s477.57
Qwen3-VL-2B-Instruct ↗NativevLLM2K102410,392.28 tok/s478.98
Qwen3-30B-A3B ↗NativevLLM2K2563,271.21 tok/s479.31
Qwen3-1.7B ↗NativevLLM2K2569,991.94 tok/s480.04
Qwen3-VL-32B-FP8 ↗FP8vLLM2K8200.43 tok/s480.55
Qwen3-14B-FP8 ↗FP8vLLM2K642,378 tok/s481.17
Qwen3-VL-4B-Instruct ↗NativevLLM2K1135.2 tok/s482.32
Qwen3-VL-2B-Instruct ↗NativevLLM2K51210,003.22 tok/s482.43
Qwen3-1.7B ↗NativevLLM2K51210,173.53 tok/s482.61
Qwen3-VL-4B-Instruct ↗NativevLLM2K644,039.3 tok/s482.73
Qwen3-14B ↗NativevLLM2K16612.19 tok/s483.77
Qwen3-14B-FP8 ↗FP8vLLM2K16777.87 tok/s483.89
Qwen3-32B-FP8 ↗FP8vLLM2K8180.29 tok/s484.8
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K2566,759.05 tok/s486.82
Qwen3-4B-FP8 ↗FP8vLLM2K2566,818.11 tok/s489.37
Qwen3-VL-32B-FP8 ↗FP8vLLM2K32662.43 tok/s489.73
Qwen3-VL-32B-FP8 ↗FP8vLLM2K128.27 tok/s489.84
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K2566,784.95 tok/s490.43
Qwen3-30B-A3B ↗NativevLLM2K1282,391.28 tok/s492.09
Qwen3-32B-FP8 ↗FP8vLLM2K128.34 tok/s492.89
Qwen3-VL-32B-FP8 ↗FP8vLLM2K16385.39 tok/s494.22
Qwen3-32B-FP8 ↗FP8vLLM2K32570.09 tok/s494.25
Qwen3-VL-32B-Instruct ↗NativevLLM2K16277.32 tok/s495.09
Qwen3-4B ↗NativevLLM2K1137.3 tok/s495.88
Qwen3-4B ↗NativevLLM2K644,529.36 tok/s495.93
Qwen3-VL-4B-FP8 ↗FP8vLLM2K2567,559 tok/s496.05
Qwen3-1.7B ↗NativevLLM2K1288,451.48 tok/s499.27
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K2565,203.97 tok/s499.77
Qwen3-4B ↗NativevLLM2K1285,700.05 tok/s502.06
Qwen3-32B ↗NativevLLM2K16233.6 tok/s503.62
Qwen3-4B-FP8 ↗FP8vLLM2K5128,143.94 tok/s505.71
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K10248,360.24 tok/s505.72
Qwen3-VL-4B-Instruct ↗NativevLLM2K1285,770.75 tok/s506.31
Qwen3-VL-4B-FP8 ↗FP8vLLM2K15368,540.41 tok/s506.34
Qwen3-VL-4B-FP8 ↗FP8vLLM2K10249,197.98 tok/s507.05
Qwen3-32B-FP8 ↗FP8vLLM2K16328.67 tok/s508.31
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K2565,651.48 tok/s508.66
Qwen3-32B-FP8 ↗FP8vLLM2K64996.85 tok/s511.7
Qwen3-VL-4B-Instruct ↗NativevLLM2K10248,525.78 tok/s512.65
Qwen3-30B-A3B ↗NativevLLM2K5124,374.82 tok/s513.07
Qwen3-VL-32B-FP8 ↗FP8vLLM2K641,181.77 tok/s514.02
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K10249,356.79 tok/s514.18
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K5128,086.71 tok/s515.9
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K2564,938.44 tok/s516.59
Qwen3-14B-FP8 ↗FP8vLLM2K1283,967.27 tok/s516.95
Qwen3-4B-Thinking-FP8 ↗FP8vLLM2K5128,078.91 tok/s517.34
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K5123,973.71 tok/s520.26
Qwen3-30B-A3B-Thinking-FP8 ↗FP8vLLM2K5126,901.04 tok/s523.06
Qwen3-8B-FP8 ↗FP8vLLM2K2565,849.1 tok/s524.95
Qwen3-4B ↗NativevLLM2K5127,662.98 tok/s526.72
Qwen3-VL-8B-Instruct ↗NativevLLM2K187.24 tok/s527.11
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K5126,561.45 tok/s527.54
Qwen3-VL-4B-FP8 ↗FP8vLLM2K5128,528.19 tok/s527.89
Qwen3-4B-Instruct-FP8 ↗FP8vLLM2K10248,326.43 tok/s528.37
Qwen3-14B ↗NativevLLM2K321,097.31 tok/s531.83
Qwen3-8B ↗NativevLLM2K188.12 tok/s532.4
Qwen3-VL-4B-Instruct ↗NativevLLM2K5127,840.44 tok/s532.55
Qwen3-VL-30B-A3B-FP8 ↗FP8vLLM2K10247,892.13 tok/s534.35
Qwen3-8B ↗NativevLLM2K5126,230.97 tok/s536.79
Qwen3-VL-8B-FP8 ↗FP8vLLM2K2566,302.19 tok/s536.93
Qwen3-8B-FP8 ↗FP8vLLM2K5127,212 tok/s537.21
Qwen3-VL-8B-FP8 ↗FP8vLLM2K10247,936.67 tok/s538.27
Qwen3-VL-8B-Instruct ↗NativevLLM2K1284,167.41 tok/s538.38
Qwen3-8B ↗NativevLLM2K1284,798.12 tok/s539.9
Qwen3-VL-8B-Instruct ↗NativevLLM2K5126,124.75 tok/s541.54
Qwen3-VL-30B-A3B-Instruct ↗NativevLLM2K10245,534.66 tok/s541.72
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K5127,491.18 tok/s543.51
Qwen3-VL-32B-Instruct ↗NativevLLM2K32460.03 tok/s543.76
Qwen3-VL-8B-Instruct ↗NativevLLM2K10246,682.77 tok/s544
Qwen3-VL-4B-Instruct ↗NativevLLM2K2566,931.45 tok/s544.78
Qwen3-4B ↗NativevLLM2K2566,245.25 tok/s545.8
Qwen3-30B-A3B-FP8 ↗FP8vLLM2K10248,866.25 tok/s545.96
Qwen3-32B ↗NativevLLM2K1281,245.43 tok/s546.58
Qwen3-32B ↗NativevLLM2K64733.6 tok/s548
Qwen3-32B-FP8 ↗FP8vLLM2K1281,748.39 tok/s548.11
Qwen3-VL-8B-Instruct ↗NativevLLM2K2565,185.84 tok/s548.18
Qwen3-VL-8B-Instruct ↗NativevLLM2K642,692.56 tok/s548.32
Qwen3-VL-32B-Instruct ↗NativevLLM2K2561,360.59 tok/s549.41
Qwen3-VL-8B-FP8 ↗FP8vLLM2K5127,344.06 tok/s551.08
Qwen3-8B ↗NativevLLM2K2564,986.94 tok/s551.14
Qwen3-14B ↗NativevLLM2K1282,725.44 tok/s551.77
Qwen3-VL-32B-Instruct ↗NativevLLM2K1281,300.44 tok/s552.11
Qwen3-8B ↗NativevLLM2K642,768.87 tok/s552.26
Qwen3-VL-32B-Instruct ↗NativevLLM2K64798.26 tok/s552.39
Qwen3-14B ↗NativevLLM2K641,594.57 tok/s552.75
Qwen3-32B ↗NativevLLM2K32437.38 tok/s553.73
Qwen3-14B ↗NativevLLM2K5124,531.19 tok/s554.2
Qwen3-VL-32B-Instruct ↗NativevLLM2K5121,731.92 tok/s556.62
Qwen3-14B ↗NativevLLM2K2563,790.12 tok/s559.06
Qwen3-VL-32B-FP8 ↗FP8vLLM2K1282,135.04 tok/s562.42
Qwen3-14B ↗NativevLLM2K151.68 tok/s564.22
Qwen3-14B-FP8 ↗FP8vLLM2K5125,808.91 tok/s569.81
Qwen3-VL-32B-FP8 ↗FP8vLLM2K10243,785.79 tok/s572.21
Qwen3-14B-FP8 ↗FP8vLLM2K2564,926.3 tok/s573.39
Qwen3-VL-32B-FP8 ↗FP8vLLM2K5123,220.42 tok/s573.65
Qwen3-VL-32B-Instruct ↗NativevLLM2K123.57 tok/s573.89
Qwen3-32B ↗NativevLLM2K123.64 tok/s575.12
Qwen3-32B-FP8 ↗FP8vLLM2K2562,702.06 tok/s575.22
Qwen3-VL-32B-FP8 ↗FP8vLLM2K2562,028.75 tok/s576.19

Original reports

Every report for this task, including quality, diagnostic and legacy runs that do not enter comparisons.
Original reports · 28 rows
ModelTest taskFormatEngineWorkloadProtocolQuality scoreTest date
Qwen3-0.6B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 4:07:56 AM
Qwen3-0.6B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 10:38:28 AM
Qwen3-0.6B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 12:40:23 AM
Qwen3-1.7B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 4:17:55 AM
Qwen3-1.7B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 1:30:18 AM
Qwen3-4B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 4:29:35 AM
Qwen3-4B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 2:47:05 AM
Qwen3-4B-Instruct-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/15/2026, 7:49:18 PM
Qwen3-4B-Thinking-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/15/2026, 7:35:40 PM
Qwen3-8B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 4:43:56 AM
Qwen3-8B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 3:02:20 AM
Qwen3-14B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 5:02:12 AM
Qwen3-14B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 3:16:47 AM
Qwen3-30B-A3B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 5:27:56 AM
Qwen3-30B-A3B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 3:59:55 AM
Qwen3-30B-A3B-Thinking-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/15/2026, 8:52:52 PM
Qwen3-32B ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 5:59:55 AM
Qwen3-32B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 3:39:20 AM
Qwen3-VL-2B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 6:19:32 AM
Qwen3-VL-2B-Instruct ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 6:10:43 AM
Qwen3-VL-4B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 6:40:36 AM
Qwen3-VL-4B-Instruct ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 6:30:05 AM
Qwen3-VL-8B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 7:13:46 AM
Qwen3-VL-8B-Instruct ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 6:54:15 AM
Qwen3-VL-30B-A3B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 7:56:40 AM
Qwen3-VL-30B-A3B-Instruct ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 7:39:24 AM
Qwen3-VL-32B-FP8 ↗Text generationFP8vLLM2KMethod unknown—1/13/2026, 8:53:29 AM
Qwen3-VL-32B-Instruct ↗Text generationNativevLLM2KMethod unknown—1/13/2026, 8:23:02 AM

Running benchmarks

0Updates automatically

No benchmarks are currently reported as running.

Methodology

Average of error-free runs for the selected recorded configuration. Missing methods and fallback runs are not ranked. Performance index = average / fastest matching hardware × 100. Value index = performance per purchase cost / best comparable value × 100. Quality is separate by dataset and protocol. Prices are dated seller listings, not completed transactions or market averages.

Updated 11/5/2026, 9:56:22 PM