| Zeile | R | tok/s | TTFT s | Prompt max | Cache | Reasoning | Tool-Anteil | Aufrufe | gleichzeitig |
|---|---|---|---|---|---|---|---|---|---|
| v3-hermes-qwen-ninfer | 3 | 38.2 | 0.09 | 55587 | 0% | 68% | 18% | 7 | 2 |
| v3-hermes-qwen-ninfer | 1 | 21.4 | 0.07 | 31335 | 0% | 73% | 9% | 16 | 2 |
| v3-hermes-qwen-ninfer | 2 | 12.2 | 0.10 | 53201 | 0% | 82% | 16% | 25 | 2 |
| v3-hermes-qwen-ninfer | 3 | 11.1 | 0.21 | 96981 | 0% | 87% | 16% | 102 | 2 |
| v3-hermes-qwen-ninfer | 4 | 22.7 | 0.10 | 48829 | 0% | 86% | 0% | 11 | 3 |
| v3-hermes-qwen-ninfer | 5 | 9.0 | 0.25 | 113786 | 0% | 83% | 20% | 82 | 2 |
| v3-hermes-qwen-ninfer | 6 | 4.3 | 0.64 | 190413 | 0% | 84% | 26% | 146 | 2 |
| v3-hermes-qwen-ninfer | 7 | 4.7 | 0.86 | 101657 | 0% | 86% | 16% | 85 | 2 |
gelb: ≥ 4 gleichzeitig offene Streams beim Aufruf — tok/s und TTFT stehen dann unter Last.
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 4/6 | —/24 | 16 | 37 | 1,371,155 | 83,153 | 2,247 | 0.91 | ok | — |
hermes · Qwen3.8-27B-ninfer · prune aus
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/5 | —/24 | 19 | local runtime without vLLM counters (ninfer :8100) | ok | fdcb2b6b | ||||
keine Proxy-Daten | ||||||||||
| 2 | 6/5 | —/24 | 38 | local runtime without vLLM counters (ninfer :8100) | ok | c28b3be0 | ||||
keine Proxy-Daten | ||||||||||
| 3 | 6/5 | —/24 | 25 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | c28b3be0 unverändert | ||||
400400400 | ||||||||||
pi · Qwen3.8-27B-ninfer
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 0/5 | —/24 | 6 | local runtime without vLLM counters (ninfer :8100) | ok | e3b0c442 | ||||
keine Proxy-Daten | ||||||||||
| 2 | 0/5 | —/24 | 6 | local runtime without vLLM counters (ninfer :8100) | ok | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 3 | 0/5 | —/24 | 10 | local runtime without vLLM counters (ninfer :8100) | ok | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
pi · Qwen3.8-27B-ninfer
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 | ||||
keine Proxy-Daten | ||||||||||
| 2 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 3 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 4 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 5 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 6 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 7 | 0/5 | —/24 | 0 | local runtime without vLLM counters (ninfer :8100) | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
pi · Qwen3.8-27B-ninfer
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 0/5 | —/24 | 0 | TimeoutError: timed out | rc=1 | e3b0c442 | ||||
keine Proxy-Daten | ||||||||||
| 2 | 0/5 | —/24 | 0 | TimeoutError: timed out | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 3 | 0/5 | —/24 | 0 | TimeoutError: timed out | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 4 | 0/5 | —/24 | 0 | TimeoutError: timed out | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
| 5 | 0/5 | —/24 | 0 | TimeoutError: timed out | rc=1 | e3b0c442 unverändert | ||||
keine Proxy-Daten | ||||||||||
hermes · Qwen3.8-27B-ninfer · ctx 235.520 · prune aus
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/5 | —/24 | 16 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | 059d33ea | ||||
400 | ||||||||||
| 2 | 6/5 | —/24 | 34 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | ea1219e8 | ||||
400 | ||||||||||
| 3 | 6/5 | —/24 | 168 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | 77c69d14 | ||||
400 | ||||||||||
| 4 | 6/5 | —/24 | 22 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | f2b27ea2 | ||||
400 | ||||||||||
| 5 | 6/5 | —/24 | 87 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | 598c7092 | ||||
400 | ||||||||||
| 6 | 6/5 | —/24 | 363 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | 04b6676c | ||||
400 | ||||||||||
| 7 | 6/5 | —/24 | 91 | local runtime without vLLM counters (set ARENA_VLLM_METRICS | ok | 3ccde97d | ||||
400 | ||||||||||
hermes · Qwen3.8-27B-Instruct · ctx 65.536 · prune 28000
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/6 | 10/24 | 47 | 53 | 1,832,743 | 187,043 | 3,529 | 0.725 | ok | 202c6658 |
| 2 | 0/6 | 0/24 | 90 | 158 | 7,457,572 | 470,521 | 2,977 | 0.733 | harness-failure | 19b6433a |
hermes · Qwen3.8-27B-Instruct · ctx 65.536 · prune 28000
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/6 | 10/24 | 31 | 34 | 1,302,874 | 77,524 | 2,280 | 0.931 | ok | — |
| 2 | 6/6 | 10/24 | 29 | TimeoutError: timed out | ok | — | ||||
| 3 | 6/6 | 14/24 | 163 | 321 | 13,668,303 | 796,792 | 2,482 | 0.808 | ok | — |
| 4 | 6/6 | 17/24 | 8 | 7 | 195,985 | 24,730 | 3,532 | 0.21 | ok | — |
| 5 | 6/6 | 17/24 | 44 | 46 | 1,763,555 | 187,673 | 4,079 | 0.571 | harness-failure | — |
| 6 | 6/6 | 17/24 | 13 | 10 | 337,524 | 46,881 | 4,688 | 0.401 | ok | — |
hermes · Qwen3.8-27B-Instruct · ctx 65.536 · prune aus
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/6 | 8/24 | 39 | 42 | 1,518,313 | 125,533 | 2,988 | 0.293 | ok | — |
| 2 | 6/6 | 11/24 | 92 | 80 | 3,602,422 | 205,915 | 2,573 | 0.268 | harness-failure | — |
| 3 | 6/6 | 12/24 | 21 | TimeoutError: timed out | harness-failure | — | ||||
| 4 | 2/6 | 10/24 | 51 | 97 | 4,418,149 | 241,605 | 2,490 | 0.73 | harness-failure | — |
| 5 | 2/6 | 10/24 | 25 | TimeoutError: timed out | harness-failure | — | ||||
| 6 | 4/6 | 10/24 | 16 | 13 | 412,456 | 59,621 | 4,586 | 0.336 | harness-failure | — |
| 7 | 3/6 | 9/24 | 6 | 4 | 83,058 | 24,898 | 6,224 | 0.0 | ok | — |
hermes · claude-sonnet-5
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2/6 | 9/24 | 14 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 2 | 2/6 | 10/24 | 7 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 3 | 2/6 | 10/24 | 25 | backend=remote: no local model, no vLLM counters | rc=-15 | — | ||||
| 4 | 2/6 | 14/24 | 14 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 5 | 2/6 | 16/24 | 18 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 6 | 2/6 | 18/24 | 15 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 7 | 2/6 | 20/24 | 32 | backend=remote: no local model, no vLLM counters | ok | — | ||||
hermes · claude-opus-5
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/6 | 12/24 | 29 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 2 | 6/6 | 12/24 | 12 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 3 | 6/6 | 12/24 | 15 | backend=remote: no local model, no vLLM counters | rc=-15 | — | ||||
| 4 | 6/6 | 17/24 | 16 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 5 | 6/6 | 20/24 | 13 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 6 | 6/6 | 24/24 | 15 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 7 | 6/6 | 24/24 | 51 | backend=remote: no local model, no vLLM counters | ok | — | ||||
hermes · claude-haiku-4-5-20251001
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2/6 | 7/24 | 10 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 2 | 6/6 | 11/24 | 12 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 3 | 6/6 | 11/24 | 14 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 4 | 6/6 | 13/24 | 14 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 5 | 6/6 | 12/24 | 29 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 6 | 4/6 | 13/24 | 4 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 7 | 1/6 | 0/24 | 4 | backend=remote: no local model, no vLLM counters | ok | — | ||||
hermes · claude-fable-5-1
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2/6 | 9/24 | 10 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 2 | 6/6 | 11/24 | 18 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 3 | 6/6 | 12/24 | 12 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 4 | 6/6 | 16/24 | 13 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 5 | 6/6 | 16/24 | 17 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 6 | 6/6 | 20/24 | 19 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 7 | 6/6 | 20/24 | 15 | backend=remote: no local model, no vLLM counters | ok | — | ||||
claude-code · claude-sonnet-5
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2/6 | 8/24 | 10 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 2 | 2/6 | 9/24 | 7 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 3 | 2/6 | 10/24 | 81 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 4 | 2/6 | 12/24 | 23 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 5 | 2/6 | 16/24 | 62 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 6 | 2/6 | 19/24 | 7 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 7 | 2/6 | 19/24 | 8 | backend=remote: no local model, no vLLM counters | ok | — | ||||
claude-code · claude-opus-5
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/6 | 10/24 | 35 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 2 | 6/6 | 10/24 | 12 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 3 | 6/6 | 11/24 | 10 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 4 | 6/6 | 15/24 | 13 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 5 | 6/6 | 19/24 | 20 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 6 | 6/6 | 23/24 | 35 | backend=remote: no local model, no vLLM counters | ok | — | ||||
| 7 | 6/6 | 22/24 | 27 | backend=remote: no local model, no vLLM counters | ok | — | ||||
claude-code · claude-fable-5-1
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 6/6 | 10/24 | 24 | backend=remote: no local model, no vLLM counters | ok | 9fc80fa0 | ||||
| 2 | 6/6 | 11/24 | 14 | backend=remote: no local model, no vLLM counters | ok | 3625c546 | ||||
| 3 | 6/6 | 12/24 | 25 | backend=remote: no local model, no vLLM counters | ok | 1f326f22 | ||||
| 4 | 6/6 | 16/24 | 14 | backend=remote: no local model, no vLLM counters | ok | caf97952 | ||||
| 5 | 6/6 | 18/24 | 28 | backend=remote: no local model, no vLLM counters | ok | d1d5cc94 | ||||
| 6 | 6/6 | 23/24 | 24 | backend=remote: no local model, no vLLM counters | ok | ddbe2634 | ||||
| 7 | 6/6 | 23/24 | 27 | backend=remote: no local model, no vLLM counters | ok | 8bd1e7f9 | ||||
hermes · Qwen3.8-27B-Instruct
| R | gate | cap | min* | req | prompt-tok | gen-tok | ø gen/req | cache* | exit | Build |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2/6 | 9/24 | 31 | 69 | 2,654,095 | 111,271 | 1,612 | 0.891 | ok | — |
| 2 | 0/6 | 0/24 | 80 | 127 | 5,691,292 | 377,816 | 2,974 | 0.776 | rc=8 | — |
ein Span je Runde, Projekt antifa-survivors-<zeile>
Span-Attribute je Runde: arena_run_id, arena_config, arena_harness, arena_model, arena_round, arena_exit_code, arena_wall_s, arena_ladder_rung, arena_capability_passed sowie ein arena_<feld> je Modell-Metrik.
* Wallzeit und Prefix-Cache tragen Einschränkungen: die eine misst die gleichzeitige Last mit (2 bis 15 parallele Runden), die andere die Serverwärme und das Alter der Session. Beide stehen hier als Kontext, nicht als Ergebnis.