Benchmark notes · 23 August 2026

Open-model inference, measured on real workers.

These are our passing runs and our failures. They are not a universal leaderboard: different runtimes, GPUs, context lengths, and request patterns answer different questions.

The results

ModelWorker and runtimeMeasured resultWorkload
DeepSeek-V4-Flash3× RTX 5090 · FreeToken30.62 end-to-end tok/s
~33.3 decode tok/s
Warmed 255-token probe
Qwen3.6 35B-A3B3× RTX 5090 · FreeToken33.76 end-to-end tok/s
~41.0 decode tok/s
Warmed 256-token probe
GLM-5.2 NVFP44× RTX PRO 6000 Blackwell Max-Q · FreeToken Triton offload7.158 end-to-end tok/s
9.235 steady decode tok/s
Warmed 255-token probe; served successfully, below the 12 tok/s qualification target
Kimi-K3 0.40B dev checkpoint1× RTX 3090 · FreeToken offload567.043 serial output tok/s
32/32 requests passed in the first concurrency ladder
Architecture proof only; not comparable to the full 2.8T model
Qwen3.6 35B-A3B8× RTX 5060 Ti · vLLM tensor parallelism 81,574 aggregate output tok/s64 concurrent requests; two successful trials
GPT-OSS 120B8× RTX 5060 Ti · vLLM tensor parallelism 81,451 aggregate output tok/s64 concurrent requests; two successful trials
Gemma4 31B NVFP48× RTX 5060 Ti · vLLM tensor parallelism 8Did not completeOut of memory during warm-up at an 8k context window

Do not turn this table into one ranking

The FreeToken rows include warmed single-request probes on multi-GPU workers and a separate concurrency test on one RTX 3090. The vLLM rows measure aggregate output across a fixed ladder on eight lower-memory GPUs. Those workloads are not interchangeable.

64successful concurrent requests reached by Qwen3.6 and GPT-OSS in the vLLM test ceiling.
8kcontext was enough to prevent Gemma4 from fitting on the 16GB-per-GPU worker.
2runtimes tested: FreeToken for hybrid MoE placement and vLLM for tensor-parallel serving.

What the Kimi run proved

FreeToken loaded the public eight-layer, 0.40B MXFP4 Kimi-K3 development checkpoint and served it through the OpenAI-compatible API on a 24GB RTX 3090. A 256-token serial request produced 255 reported output tokens in 0.450 seconds, or 567.043 tokens/s. The first ladder completed all 32 simultaneous requests.

This proves the new KDA/MLA, latent-MoE, SiTU, MXFP4 loading, batching, and API path. It says nothing about how fast the full 2.8T Kimi-K3 model will run.

During a sustained follow-up, the live-swapped NVIDIA driver lost the passed-through GPU on the second 16-request sample. We kept the completed samples and recorded the failure. We did not turn the earlier burst peak into a sustained-throughput claim.

Review the FreeToken implementation and the saved benchmark artifacts.

What this changes for routing

Parameter count is not enough. Placement needs the weight format, active memory footprint, context limit, runtime, and expected concurrency.

A cheap community worker can be right for one workload and unable to load another. We keep raw results and failed runs so that decision can be checked later.

Qwen3.8 27B remains a separate candidate for a 24GB RTX 3090. Its fit has been assessed, but it has no TEKIZ.AI benchmark result yet.

Method notes

The vLLM runs used fixed 128-token completions, temperature 0, a fixed prompt seed, tensor parallelism across eight GPUs, and two trials at each concurrency level through 64. FreeToken requests streamed over the OpenAI-compatible API. The Kimi run ignored EOS to exercise the requested token budget. We saved metrics, not generated text or credentials. These are throughput tests, not quality scores or maximum-capacity claims.

Run a community provider or start with the API quickstart.