{"models":[{"id":"Intel/Qwen3.5-397B-A17B-int4-AutoRound","name":"Qwen 3.5 397B","size":"397B (17B aktiv)","category":"Frontier","status":"loaded","availability":"permanent","featured":true,"description":"Alibabas größtes Open-Source-Modell. 397B Parameter, 17B aktiv (MoE). Frontier-Qualität für anspruchsvolle Aufgaben."},{"id":"qwen3.6-35b","name":"Qwen 3.6 35B","size":"35B (3B aktiv)","category":"Allgemein","status":"loaded","availability":"permanent","featured":true,"description":"Hybrid-MoE (Gated DeltaNet + Attention) mit 35B Parametern, 3B aktiv. Multimodal (Text/Bild/Video), Tools, Reasoning. Läuft auf spark-3 unter vLLM, dauerhaft warm. Hinter den Aliasen mid, vision und coder."},{"id":"lfm2.5-8b","name":"LFM 2.5 8B","size":"8B","category":"Schnell","status":"loaded","availability":"permanent","featured":false,"description":"Liquid Foundation Model 2.5 (Liquid AI). Spezialisiert auf Klassifikation, Routing und Intent-Checks: A/B/C-Entscheidungen via response_format=json_schema mit enum, ~255ms pro Entscheidung. NICHT für Wissensfragen oder Code-Generierung. Hinter dem Alias fast. Warm auf spark-3."},{"id":"qwen3-coder-next","name":"Qwen3 Coder Next 80B","size":"80B","category":"Code","status":"loaded","availability":"on_demand","featured":false,"description":"Spezialisiert auf agentic Coding, SWE-Bench 69.6%."},{"id":"qwen3.5:35b-a3b","name":"Qwen 3.5 35B","size":"35B (3B aktiv)","category":"Allgemein","status":"loaded","availability":"on_demand","featured":false,"description":"MoE-Modell mit 35B Parametern und 3B aktiven. Gute Balance aus Geschwindigkeit und Qualität für alltägliche Aufgaben."},{"id":"qwen3.6:35b-a3b-q8_0","name":"Qwen 3.6 35B (Ollama)","size":"35B (3B aktiv)","category":"Allgemein","status":"loaded","availability":"on_demand","featured":false,"description":"Ollama-Variante (Q8_0). Direkt per Modell-ID erreichbar; die kuratierten Aliase mid/vision/coder zeigen auf die vLLM-Variante qwen3.6-35b."},{"id":"ornith:35b","name":"Ornith 35B","size":"35B (qwen35moe, Q4_K_M)","category":"Code","status":"loaded","availability":"on_demand","featured":false,"description":"Self-improving agentic Coding-Modell (deep-reinforce.com, MIT). RL-optimiert für Terminal-Bench, SWE-Bench, NL2Repo, OpenClaw. On-demand auf spark-4."},{"id":"nemotron3:33b","name":"Nemotron 3 Omni","size":"33B","category":"Multimodal","status":"loaded","availability":"on_demand","featured":false,"description":"NVIDIAs Omni-Modell. Text, Bild, Audio (Beta). Tools und Reasoning."},{"id":"deepseek-r1:32b","name":"DeepSeek R1 32B","size":"32B","category":"Reasoning","status":"loaded","availability":"on_demand","featured":false,"description":"DeepSeeks Reasoning-Modell mit 32B Parametern. Stark bei Logik und Mathematik."},{"id":"qwen3-coder:30b","name":"Qwen 3 Coder 30B","size":"30B","category":"Code","status":"loaded","availability":"on_demand","featured":false,"description":"Qwen Coding-Modell mit 30B Parametern. Für komplexe Code-Aufgaben."},{"id":"nemotron-3-nano:30b","name":"Nemotron 3 Nano","size":"30B (3.5B aktiv)","category":"Allgemein","status":"loaded","availability":"on_demand","featured":false,"description":"NVIDIAs Hybrid-MoE (Mamba-2 + Attention). Effizient bei langem Kontext."},{"id":"granite4.1:30b","name":"Granite 4.1","size":"30B","category":"Allgemein","status":"loaded","availability":"on_demand","featured":false,"description":"IBMs Granite 4.1 Modell. Open-Source, optimiert für Enterprise-Aufgaben und Long-Context."},{"id":"qwen3.6:27b","name":"Qwen 3.6 27B","size":"27B","category":"Allgemein","status":"loaded","availability":"on_demand","featured":false,"description":"Qwen-Dense-Modell mit Vision-Support. Solides Allround-Modell für Chat und Bildanalyse."},{"id":"gemma4:26b","name":"Gemma 4 26B","size":"26B","category":"Allgemein","status":"loaded","availability":"on_demand","featured":false,"description":"Googles offenes 26B Modell. Vielseitig und effizient, ~34 t/s auf Ollama."},{"id":"phi4:14b","name":"Phi 4 14B","size":"14B","category":"Schnell","status":"loaded","availability":"on_demand","featured":false,"description":"Microsofts kompaktes 14B Reasoning-Modell. Optimiert für Effizienz."},{"id":"qwen3.5:9b","name":"Qwen 3.5 9B","size":"9B","category":"Schnell","status":"loaded","availability":"on_demand","featured":false,"description":"Kompaktes Modell für einfache Aufgaben und als Checkin-Modell für die Tool-Vorprüfung. Minimale Latenz."},{"id":"qwen3-embedding:4b","name":"Qwen3 Embedding 4B","size":"4B","category":"Embedding","status":"loaded","availability":"on_demand","featured":false,"description":"Embedding-Modell für semantische Suche und RAG. 4B Parameter, stärker als nomic-embed-text bei mehrsprachigen und längeren Passagen. Über den embed-Alias erreichbar."},{"id":"nemotron-cascade-2","name":"Nemotron Cascade 2","size":"","category":"Reasoning","status":"loaded","availability":"on_demand","featured":false,"description":"NVIDIAs kaskadierendes Reasoning-Modell. Gold-Medaille IMO/IOI/ICPC."},{"id":"glm-4.7-flash","name":"GLM-4.7 Flash","size":"","category":"Schnell","status":"loaded","availability":"on_demand","featured":false,"description":"Schnelles Modell von Zhipu AI mit MLA. Niedrige Latenz."}],"last_updated":"2026-07-23T09:07:38.877400+00:00","performance":[{"model":"lfm2.5-8b","tok_per_sec":60.07281553235105,"request_count":3,"total_tokens_out":300,"ttft_ms":112,"ttft_warm_ms":84,"computed_at":"2026-07-23 03:01:31","display_name":"LFM 2.5 8B"},{"model":"qwen3.6-35b","tok_per_sec":53.45572354340694,"request_count":3,"total_tokens_out":300,"ttft_ms":149,"ttft_warm_ms":121,"computed_at":"2026-07-23 03:00:54","display_name":"Qwen 3.6 35B"},{"model":"nemotron-3-nano:30b","tok_per_sec":40.54624806351774,"request_count":1,"total_tokens_out":289,"ttft_ms":1093,"ttft_warm_ms":1093,"computed_at":"2026-07-23 03:35:13","display_name":"Nemotron 3 Nano"},{"model":"nemotron-cascade-2","tok_per_sec":40.19704434087668,"request_count":2,"total_tokens_out":372,"ttft_ms":2196,"ttft_warm_ms":1075,"computed_at":"2026-07-23 04:09:03","display_name":"Nemotron Cascade 2"},{"model":"ornith:35b","tok_per_sec":40.102484126299736,"request_count":3,"total_tokens_out":3084,"ttft_ms":21396,"ttft_warm_ms":1688,"computed_at":"2026-07-23 03:09:28","display_name":"Ornith 35B"},{"model":"qwen3.5:35b-a3b","tok_per_sec":39.94252873050124,"request_count":2,"total_tokens_out":246,"ttft_ms":26521,"ttft_warm_ms":1496,"computed_at":"2026-07-23 03:06:30","display_name":"Qwen 3.5 35B"},{"model":"nemotron3:33b","tok_per_sec":39.92210321015314,"request_count":3,"total_tokens_out":3196,"ttft_ms":21871,"ttft_warm_ms":833,"computed_at":"2026-07-22 03:19:52","display_name":"Nemotron 3 Omni"},{"model":"qwen3-coder-next","tok_per_sec":27.492363234028147,"request_count":3,"total_tokens_out":300,"ttft_ms":1674,"ttft_warm_ms":824,"computed_at":"2026-07-23 03:02:21","display_name":"Qwen3 Coder Next 80B"},{"model":"Intel/Qwen3.5-397B-A17B-int4-AutoRound","tok_per_sec":26.456440415470976,"request_count":3,"total_tokens_out":300,"ttft_ms":239,"ttft_warm_ms":211,"computed_at":"2026-07-23 03:00:16","display_name":"Qwen 3.5 397B"},{"model":"gemma4:26b","tok_per_sec":21.083937941667212,"request_count":3,"total_tokens_out":6681,"ttft_ms":14672,"ttft_warm_ms":2719,"computed_at":"2026-07-23 03:58:22","display_name":"Gemma 4 26B"},{"model":"qwen3.5:9b","tok_per_sec":19.734137321057784,"request_count":3,"total_tokens_out":669,"ttft_ms":56446,"ttft_warm_ms":1799,"computed_at":"2026-07-23 04:01:49","display_name":"Qwen 3.5 9B"},{"model":"phi4:14b","tok_per_sec":15.036452007446629,"request_count":3,"total_tokens_out":300,"ttft_ms":12230,"ttft_warm_ms":413,"computed_at":"2026-07-23 03:59:33","display_name":"Phi 4 14B"},{"model":"qwen3.6:27b","tok_per_sec":7.93123100309003,"request_count":1,"total_tokens_out":502,"ttft_ms":1996,"ttft_warm_ms":1996,"computed_at":"2026-07-19 03:52:29","display_name":"Qwen 3.6 27B"},{"model":"granite4.1:30b","tok_per_sec":7.560137456657544,"request_count":3,"total_tokens_out":300,"ttft_ms":20854,"ttft_warm_ms":612,"computed_at":"2026-07-23 03:36:58","display_name":"Granite 4.1"},{"model":"qwen3-coder:30b","tok_per_sec":1.3927576601957439,"request_count":3,"total_tokens_out":300,"ttft_ms":46991,"ttft_warm_ms":1332,"computed_at":"2026-07-23 03:25:34","display_name":"Qwen 3 Coder 30B"}]}