DeepInfra: Intelligenz, Leistung und Preise der Modelle
Diese Analyse soll Sie bei der Auswahl des besten von DeepInfra angebotenen Modells für Ihren Anwendungsfall unterstützen.
Höchste Intelligenz
MiMo-V2.6-Pro
DeepSeek V4.1 Flash 1017 (max)
GLM-5.3 (max)GLM-5.3-Flash
Ling 3.1 FlashIntelligence Index
Insgesamt 106 Modelle
Am schnellsten
Inkling (xhigh) (FP8)Inkling Small
Nemotron 3 Nano (non-reasoning)Nemotron 3 NanoNemotron 3.5 Lightning (NVFP4)Ausgabegeschwindigkeit
Insgesamt 106 Modelle
Niedrigster Preis
Llama 3.1 8B (Turbo, FP8)Llama 3.1 8B
Granite 4.2 3B
Gemma 4 E4BGemma 4 E4B (non-reasoning)Mischpreis (pro 1 Mio. Tokens)
Insgesamt 106 Modelle
DeepInfra bietet 106 Modelle mit jeweils unterschiedlichen Merkmalen bei Intelligenz, Leistung und Preis an. Nachfolgend werden die wichtigsten Metriken der Modelle verglichen.
- Bei der Intelligenz sind MiMo-V2.6-Pro (46), DeepSeek V4.1 Flash 1017 (max) (45) und GLM-5.3 (max) (45) die führenden Modelle von DeepInfra.
- Bei der Ausgabegeschwindigkeit sind Inkling (xhigh) (FP8) (328 t/s), Inkling Small (301 t/s) und Nemotron 3 Nano (non-reasoning) (227 t/s) am schnellsten. Die Geschwindigkeit unterscheidet sich deutlich zwischen den Modellen; der Unterschied zwischen dem schnellsten und dem langsamsten beträgt 52 %.
- Bei der Latenz bieten Nemotron 3 Nano (non-reasoning) (0.51 s), Qwen3 30B (non-reasoning) (FP8) (0.58 s) und NVIDIA Nemotron Nano 9B V2 (non-reasoning) (0.58 s) die kürzeste Zeit bis zum ersten Antworttoken.
- Beim Preis bieten Llama 3.1 8B (Turbo, FP8) ($0.02), Llama 3.1 8B ($0.02) und Granite 4.2 3B ($0.02) die niedrigsten Mischpreise pro 1 Mio. Tokens.
- Bei der Größe des Kontextfensters unterstützen GLM-5.2 (max) (FP4) (1M), DeepSeek V4 Pro (max) (FP4) (1M) und DeepSeek V4 Flash (max) (FP4) (1M) die größten Kontextfenster von DeepInfra.
Intelligenzevaluationen
Artificial Analysis Intelligence Index
Evaluations
Agentic knowledge work, Elo
Knowledge
1 - hallucination rate
Agentic SaaS workflows
Agentic real-world work tasks, Elo
Reasoning & knowledge
Professional document reasoning, All-pass
Agentic coding & terminal use
Long context reasoning
Coding
Physics reasoning
Agentic scientific research workflows in a terminal
Agentic software engineering in open-source repositories
Quantitative analysis on spreadsheets & documents
Legal agentic work, Hallucination-Gated All-Pass Rate
Agentic business operations
Agentic interactive 3D experiences, Elo
Product and brand videos designed in code, Elo
Kubernetes incident root-cause analysis
Medical long context reasoning
Visual reasoning
Rich, varied, well-crafted language
Well-structured, scannable, answer-first
Truthful, acknowledges uncertainty, non-sycophantic
Use of filler and generic language
Intelligence Index vs. Preis
Kontextfenster
Kontextfenster
Preise
Intelligence Index vs. Preis
Leistungsübersicht
Ausgabegeschwindigkeit vs. Preis
Geschwindigkeit
Gemessen anhand der Ausgabegeschwindigkeit (Tokens pro Sekunde)
Ausgabegeschwindigkeit
Latenz
Gemessen anhand der Zeit (Sekunden) bis zum ersten Token
Latenz: Zeit bis zum ersten Antworttoken
Cache-Verhalten
Cache-Trefferquote
Kosten pro Aufgabe vs. Cache-Trefferquote
Ende-zu-Ende-Antwortzeit
Seconds to output 500 tokens, calculated based on time to first token, 'thinking' time for reasoning models, and output speed
Ende-zu-Ende-Antwortzeit vs. Preis
Weitere Analyse | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
MiMo-V2.6-Pro | 1M | Offen | 46 | $0.26 | 46 | 1.56 | 55.85 | 43.43 | |||
DeepSeek V4.1 Flash 1017 (max) | 1M | Offen | 45 | $0.08 | 67 | 0.68 | 37.84 | 29.73 | |||
GLM-5.3 (max) | 1.05M | Offen | 45 | $0.97 | 130 | 0.93 | 20.13 | 15.36 | |||
GLM-5.3-Flash | 1M | Offen | 42 | $0.56 | 96 | 0.76 | 26.70 | 20.74 | |||
Ling 3.1 Flash | 262k | Offen | 41 | $0.19 | 18 | 1.04 | 141.47 | 112.34 | |||
 Qwen3.8 2.4T A95B | 262k | Offen | 40 | $2.16 | 108 | 0.97 | 24.05 | 18.46 | |||
DeepSeek V4.1 Flash (max) | 1M | Offen | 39 | $0.15 | 126 | 0.68 | 20.55 | 15.90 | |||
MiMo-V2.6-Flash | 1M | Offen | 38 | $0.04 | 25 | 9.12 | 109.70 | 80.47 | |||
DeepSeek V4 Pro 0813 (max) | 1M | Offen | 36 | $1.00 | 93 | 0.99 | 27.79 | 21.44 | |||
DeepSeek V4 Flash Vision (max) | 1M | Proprietär | 35 | $0.51 | 207 | 0.86 | 12.97 | 9.68 | |||
DeepSeek V4 Flash 0731 (max) | 1M | Offen | 34 | $0.09 | 66 | 0.79 | 38.56 | 30.22 | |||
GLM-5.2 (max) (FP4) | 1.05M | Offen | 34 | $0.37 | 70 | 1.12 | 36.89 | 28.62 | |||
Qwen3.8 27B (xhigh) | 262k | Offen | 34 | $0.52 | 33 | 1.19 | 77.59 | 61.12 | |||
DeepSeek V4 Pro (max) (FP4) | 1.05M | Offen | 30 | $0.87 | 81 | 1.12 | 61.05 | 53.78 | |||
DeepSeek V4 Pro (high) (FP4) | 65.5k | Offen | 30* | -- | -- | -- | -- | -- | |||
 MiniMax-M3 | 524k | Offen | 29 | $0.44 | 25 | 0.87 | 102.58 | 81.37 | |||
GLM-5 (FP4) | 203k | Offen | 28* | -- | 80 | 1.04 | 46.20 | 38.90 | |||
 Kimi K2.6 (FP4) | 262k | Offen | 27 | $0.51 | 62 | 0.99 | 81.07 | 72.00 | |||
GLM-5.1 (FP4) | 203k | Offen | 26 | $0.58 | 102 | 0.91 | 42.88 | 37.08 | |||
MiMo-V2.5-Pro | 65.5k | Offen | 26 | $0.19 | 41 | 1.89 | 63.01 | 48.90 | |||
Kimi K2.7 Code | 262k | Offen | 26 | $1.37 | 88 | 1.36 | 32.50 | 25.44 | |||
Inkling Small | 524k | Offen | 26 | $0.12 | 301 | 0.70 | 9.01 | 6.65 | |||
 Hy3 (FP8) | 262k | Offen | 25 | $0.07 | 47 | 1.70 | 55.32 | 42.90 | |||
Inkling (xhigh) (FP8) | 131k | Offen | 25 | $0.56 | 328 | 0.69 | 8.31 | 6.10 | |||
DeepSeek V4 Flash (high) (FP4) | 1.05M | Offen | 24 | $0.12 | -- | -- | -- | -- | |||
GLM-5.1 (non-reasoning) (FP4) | 203k | Offen | 24* | -- | 117 | 1.05 | 5.34 | -- | |||
DeepSeek V4 Flash (max) (FP4) | 1.05M | Offen | 24 | $0.07 | 32 | 0.95 | 190.16 | 173.73 | |||
Kimi K2.6 (non-reasoning) (FP4) | 262k | Offen | 24* | -- | 31 | 0.94 | 17.09 | -- | |||
Kimi K2.5 | 262k | Offen | 23* | -- | 47 | 1.06 | 74.83 | 63.13 | |||
Nemotron 3 Ultra | 262k | Offen | 23 | $0.29 | 124 | 1.23 | 23.54 | 18.30 | |||
Nemotron 3 Ultra BF16 | 262k | Offen | 23 | $0.66 | 91 | 1.42 | 31.84 | 24.94 | |||
Qwen3.5 27B (FP8) | 262k | Offen | 23* | -- | 59 | 1.08 | 43.32 | 33.79 | |||
MiniMax-M2.5 (FP8) | 197k | Offen | 23* | -- | 26 | 0.82 | 97.60 | 77.42 | |||
GLM-4.7 (FP4) | 203k | Offen | 22* | -- | 24 | 1.13 | 104.46 | 82.67 | |||
MiMo-V2.5 | 262k | Offen | 22 | $0.02 | 22 | 8.26 | 122.16 | 91.12 | |||
GLM-5 (non-reasoning) (FP8) | 203k | Offen | 22* | -- | 76 | 0.95 | 7.53 | -- | |||
DeepSeek V3.2 (FP4) | 164k | Offen | 21* | -- | 32 | 0.99 | 80.24 | 63.40 | |||
Qwen3.5 397B A17B (non-reasoning) (FP8) | 262k | Offen | 21* | -- | 49 | 1.14 | 11.40 | -- | |||
Qwen3.6 27B FP8 | 262k | Offen | 21 | $0.37 | 56 | 1.09 | 112.31 | 102.22 | |||
Ling 3.0 Flash | 131k | Offen | 20 | $0.02 | 62 | 63.17 | 103.76 | 32.47 | |||
Qwen3.6 27B (non-reasoning) FP8 | 262k | Offen | 20* | -- | 59 | 1.17 | 9.65 | -- | |||
 Step 3.7 Flash | 256k | Offen | 19* | -- | 24 | 10.48 | 114.78 | 83.44 | |||
Qwen3.5 27B (non-reasoning) FP8 | 262k | Offen | 19* | -- | 55 | 1.04 | 10.09 | -- | |||
Qwen3.5 35B A3B (FP8) | 262k | Offen | 19* | -- | 137 | 0.69 | 18.96 | 14.62 | |||
GLM-4.6 (FP4) | 203k | Offen | 19* | -- | 23 | 1.23 | 107.63 | 85.12 | |||
Qwen3.5 397B A17B (FP8) | 262k | Offen | 18 | $0.25 | 49 | 1.27 | 76.90 | 65.37 | |||
MiMo-V2.5-Pro (non-reasoning) | 65.5k | Offen | 18* | -- | 41 | 1.45 | 13.74 | -- | |||
Qwen3.6 35B A3B (FP8) | 262k | Offen | 18 | $0.14 | 111 | 0.85 | 53.74 | 48.40 | |||
Qwen3.5 122B A10B (non-reasoning) (FP4) | 262k | Offen | 18* | -- | 41 | 1.17 | 13.26 | -- | |||
Muse Glimmer (high) | 131k | Offen | 17 | $0.05 | 124 | 0.98 | 21.16 | 16.15 | |||
GLM-4.7 (non-reasoning) (FP4) | 203k | Offen | 17* | -- | 28 | 1.08 | 19.21 | -- | |||
Gemma 4 26B A4B (FP8) | 262k | Offen | 17* | -- | 75 | 0.72 | 34.06 | 26.67 | |||
DeepSeek V3.2 (non-reasoning) | 164k | Offen | 16* | -- | 34 | 1.15 | 15.88 | -- | |||
Qwen3.5 122B A10B (FP4) | 262k | Offen | 16 | $0.24 | 46 | 1.15 | 54.92 | 43.01 | |||
Qwen3.6 35B A3B (non-reasoning) (FP8) | 262k | Offen | 15* | -- | 120 | 0.77 | 4.93 | -- | |||
Qwen3.5 35B A3B (non-reasoning) FP8 | 262k | Offen | 15* | -- | 145 | 0.77 | 4.22 | -- | |||
GLM-4.7-Flash | 203k | Offen | 15* | -- | 91 | 0.86 | 28.29 | 21.94 | |||
Gemma 4 31B | 262k | Offen | 15 | $0.15 | 24 | 1.63 | 93.78 | 71.54 | |||
DeepSeek V3.1 Terminus (non-reasoning) (FP4) | 164k | Offen | 14* | -- | 75 | 0.97 | 7.66 | -- | |||
Gemma 4 31B (non-reasoning) (FP8) | 262k | Offen | 14* | -- | 24 | 1.29 | 21.79 | -- | |||
DeepSeek V3.1 (non-reasoning) (FP4) | 164k | Offen | 14* | -- | 10 | 1.03 | 50.24 | -- | |||
Gemma 4 26B A4B (non-reasoning) (FP8) | 262k | Offen | 13* | -- | 68 | 0.72 | 8.04 | -- | |||
Qwen3.5 4B (FP8) | 262k | Offen | 13* | -- | 18 | 0.89 | 138.85 | 110.37 | |||
DeepSeek R1 0528 | 164k | Offen | 13* | -- | 39 | 0.93 | 65.51 | 51.67 | |||
Nemotron 3.5 Lightning (NVFP4) | 262k | Offen | 13 | $0.09 | 216 | 0.53 | 12.09 | 9.24 | |||
Nemotron 3 Super | 262k | Offen | 13 | $0.48 | 70 | 12.13 | 47.97 | 28.67 | |||
Qwen3 235B A22B 2507 (FP8) | 262k | Offen | 13 | $0.08 | 125 | 0.76 | 20.83 | 16.06 | |||
Qwen3 235B 2507 (FP8) | 262k | Offen | 12* | -- | 14 | 0.73 | 36.64 | -- | |||
Qwen3 Coder 480B (Turbo, FP4) | 262k | Offen | 12* | -- | 61 | 0.70 | 8.86 | -- | |||
 gpt-oss-120b (high) (Turbo) | 131k | Offen | 12 | $0.11 | 45 | 0.68 | 55.69 | 44.01 | |||
gpt-oss-120b (high) | 131k | Offen | 12 | $0.03 | 47 | 0.66 | 53.49 | 42.26 | |||
Granite 4.2 8B | 131k | Offen | 11 | $0.02 | 49 | 0.89 | 52.33 | 41.16 | |||
Qwen3.5 4B (non-reasoning) FP8 | 262k | Offen | 11* | -- | 19 | 0.83 | 26.59 | -- | |||
gpt-oss-120b (low) | 131k | Offen | 10* | -- | 45 | 0.66 | 55.68 | 44.02 | |||
DeepSeek V3 0324 (FP4) | 164k | Offen | 10 | $0.02 | 136 | 0.75 | 4.43 | -- | |||
Qwen3 Next 80B A3B | 262k | Offen | 10* | -- | 128 | 0.73 | 4.64 | -- | |||
Llama 4 Maverick (FP8) | 1.05M | Offen | 9 | $0.03 | 47 | 0.79 | 11.47 | -- | |||
Granite 4.2 3B | 131k | Offen | 9 | $0.01 | 49 | 0.85 | 52.18 | 41.06 | |||
gpt-oss-20b (high) | 131k | Offen | 9 | $0.01 | 120 | 0.52 | 21.27 | 16.60 | |||
Gemma 4 E4B | 131k | Offen | 9 | $0.01 | 47 | 1.09 | 54.46 | 42.70 | |||
Nemotron 3 Nano | 262k | Offen | 9 | $0.02 | 221 | 0.52 | 11.82 | 9.04 | |||
DeepSeek V3 (Dec) | 164k | Offen | 8 | $0.02 | 25 | 0.75 | 20.72 | -- | |||
DeepSeek R1 Distill Llama 70B | 131k | Offen | 8* | -- | -- | -- | -- | -- | |||
Qwen2.5 72B | 32.8k | Offen | 8* | -- | 25 | 2.67 | 22.53 | -- | |||
Llama 3.3 70B (Turbo, FP8) | 131k | Offen | 8* | -- | 17 | 2.36 | 32.12 | -- | |||
Qwen3 30B (FP8) | 41k | Offen | 8* | -- | 109 | 0.58 | 23.45 | 18.30 | |||
Gemma 4 E4B (non-reasoning) | 131k | Offen | 7* | -- | 45 | 0.86 | 11.88 | -- | |||
NVIDIA Nemotron Nano 9B V2 | 131k | Offen | 7* | -- | 181 | 0.55 | 14.39 | 11.07 | |||
Qwen3 32B (non-reasoning) (FP8) | 41k | Offen | 7* | -- | 29 | 1.42 | 18.45 | -- | |||
Qwen3 32B (FP8) | 41k | Offen | 7 | -- | 31 | 1.72 | 81.43 | 63.77 | |||
 Mistral Small 3.1 | 128k | Offen | 7 | $0.02 | 48 | 1.01 | 11.39 | -- | |||
Mistral Small 3.2 (FP8) | 128k | Offen | 7 | $0.11 | 47 | 0.93 | 11.48 | -- | |||
Llama 3.1 8B (Turbo, FP8) | 131k | Offen | 7* | -- | 33 | 0.92 | 16.11 | -- | |||
Llama 3.1 8B | 131k | Offen | 7* | -- | 33 | 0.95 | 16.25 | -- | |||
Nemotron 3 Nano (non-reasoning) | 262k | Offen | 7* | -- | 227 | 0.51 | 2.72 | -- | |||
NVIDIA Nemotron Nano 9B V2 (non-reasoning) | 131k | Offen | 7* | -- | 193 | 0.58 | 3.17 | -- | |||
Qwen3 14B (non-reasoning) (FP8) | 41k | Offen | 7* | -- | 66 | 0.87 | 8.46 | -- | |||
Mistral Small 3 | 32.8k | Offen | 7* | -- | 55 | 0.90 | 10.06 | -- | |||
Qwen3 30B (non-reasoning) (FP8) | 41k | Offen | 7* | -- | 107 | 0.58 | 5.26 | -- | |||
Llama 3.1 70B | 131k | Offen | 7* | -- | 41 | 2.00 | 14.30 | -- | |||
Llama 3.1 70B (Turbo, FP8) | 131k | Offen | 7* | -- | 39 | 2.03 | 14.86 | -- | |||
Llama 4 Scout | 131k | Offen | 6 | $0.06 | 22 | 1.38 | 23.62 | -- | |||
Qwen3 14B (FP8) | 32.8k | Offen | 6 | -- | 70 | 0.90 | 36.71 | 28.65 | |||
 Hermes 3 - Llama-3.1 70B | 131k | Offen | 6* | -- | 31 | 2.06 | 17.93 | -- | |||
 Phi-4 | 16.4k | Offen | 6* | -- | 93 | 1.06 | 6.42 | -- | |||
Llama 3.2 11B (Vision) | 131k | Offen | 5* | -- | 30 | 1.28 | 17.77 | -- | |||
Gemma 3 27B | 131k | Offen | 5 | $0.16 | 22 | 1.45 | 24.46 | -- | |||
Gemma 3 4B | 131k | Offen | 5* | -- | 91 | 0.59 | 6.10 | -- | |||
Llama 3 8B | 8.19k | Offen | 5* | -- | -- | -- | -- | -- | |||
Gemma 3 12B | 131k | Offen | 4 | $0.13 | 23 | 1.22 | 22.87 | -- | |||
Wichtige Definitionen
Häufig gestellte Fragen
Häufige Fragen zu DeepInfra
DeepInfra bietet 106 von uns erfasste Modelle an: MiMo-V2.6-Pro, DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), GLM-5.3-Flash, Ling 3.1 Flash, Qwen3.8 2.4T A95B, DeepSeek V4.1 Flash (max), MiMo-V2.6-Flash, DeepSeek V4 Pro 0813 (max), DeepSeek V4 Flash Vision (max), DeepSeek V4 Flash 0731 (max), GLM-5.2 (max) (FP4), Qwen3.8 27B (xhigh), DeepSeek V4 Pro (max) (FP4), MiniMax-M3, GLM-5 (FP4), Kimi K2.6 (FP4), GLM-5.1 (FP4), MiMo-V2.5-Pro, Kimi K2.7 Code, Inkling Small, Hy3 (FP8), Inkling (xhigh) (FP8), GLM-5.1 (non-reasoning) (FP4), DeepSeek V4 Flash (max) (FP4), Kimi K2.6 (non-reasoning) (FP4), Kimi K2.5, Nemotron 3 Ultra, Nemotron 3 Ultra BF16, Qwen3.5 27B (FP8), MiniMax-M2.5 (FP8), GLM-4.7 (FP4), MiMo-V2.5, GLM-5 (non-reasoning) (FP8), DeepSeek V3.2 (FP4), Qwen3.5 397B A17B (non-reasoning) (FP8), Qwen3.6 27B FP8, Ling 3.0 Flash, Qwen3.6 27B (non-reasoning) FP8, Step 3.7 Flash, Qwen3.5 27B (non-reasoning) FP8, Qwen3.5 35B A3B (FP8), GLM-4.6 (FP4), Qwen3.5 397B A17B (FP8), MiMo-V2.5-Pro (non-reasoning), Qwen3.6 35B A3B (FP8), Qwen3.5 122B A10B (non-reasoning) (FP4), Muse Glimmer (high), GLM-4.7 (non-reasoning) (FP4), Gemma 4 26B A4B (FP8), DeepSeek V3.2 (non-reasoning), Qwen3.5 122B A10B (FP4), Qwen3.6 35B A3B (non-reasoning) (FP8), Qwen3.5 35B A3B (non-reasoning) FP8, GLM-4.7-Flash, Gemma 4 31B, DeepSeek V3.1 Terminus (non-reasoning) (FP4), Gemma 4 31B (non-reasoning) (FP8), DeepSeek V3.1 (non-reasoning) (FP4), Gemma 4 26B A4B (non-reasoning) (FP8), Qwen3.5 4B (FP8), DeepSeek R1 0528, Nemotron 3.5 Lightning (NVFP4), Nemotron 3 Super, Qwen3 235B A22B 2507 (FP8), Qwen3 235B 2507 (FP8), Qwen3 Coder 480B (Turbo, FP4), gpt-oss-120b (high) (Turbo), gpt-oss-120b (high), Granite 4.2 8B, Qwen3.5 4B (non-reasoning) FP8, gpt-oss-120b (low), DeepSeek V3 0324 (FP4), Qwen3 Next 80B A3B, Llama 4 Maverick (FP8), Granite 4.2 3B, gpt-oss-20b (high), Gemma 4 E4B, Nemotron 3 Nano, DeepSeek V3 (Dec), Qwen2.5 72B, Llama 3.3 70B (Turbo, FP8), Qwen3 30B (FP8), Gemma 4 E4B (non-reasoning), NVIDIA Nemotron Nano 9B V2, Qwen3 32B (non-reasoning) (FP8), Qwen3 32B (FP8), Mistral Small 3.1, Mistral Small 3.2 (FP8), Llama 3.1 8B (Turbo, FP8), Llama 3.1 8B, Nemotron 3 Nano (non-reasoning), NVIDIA Nemotron Nano 9B V2 (non-reasoning), Qwen3 14B (non-reasoning) (FP8), Mistral Small 3, Qwen3 30B (non-reasoning) (FP8), Llama 3.1 70B, Llama 3.1 70B (Turbo, FP8), Llama 4 Scout, Qwen3 14B (FP8), Hermes 3 - Llama-3.1 70B, Phi-4, Llama 3.2 11B (Vision), Gemma 3 27B, Gemma 3 4B und Gemma 3 12B.
Das intelligenteste bei DeepInfra verfügbare Modell ist MiMo-V2.6-Pro mit einem Wert von 46 im Intelligence Index.
Gemessen an der Ausgabegeschwindigkeit ist Inkling (xhigh) (FP8) mit 328.0 Tokens pro Sekunde das schnellste Modell von DeepInfra.
Das Modell von DeepInfra mit der kürzesten Zeit bis zum ersten Antworttoken ist Nemotron 3 Nano (non-reasoning) mit 0.51 s. Eine niedrigere Latenz bedeutet eine schnellere erste Antwort.
Gemessen am Mischpreis ist Llama 3.1 8B (Turbo, FP8) mit $0.02 pro 1 Mio. Tokens (Verhältnis von Cache-Treffern, Eingabe und Ausgabe: 7:2:1) das günstigste Modell von DeepInfra.
Die Preise der Modelle von DeepInfra unterscheiden sich bis zum 52-Fachen: von $0.02 pro 1 Mio. Tokens für Llama 3.1 8B (Turbo, FP8) bis $1.14 pro 1 Mio. Tokens für Qwen3.8 2.4T A95B.
Ja, DeepInfra bietet eine OpenAI-kompatible API. Das erleichtert den Wechsel von OpenAI und die Verwendung bestehender Integrationen mit dem OpenAI SDK.
Ja, alle 106 Modelle von DeepInfra unterstützen den JSON-Modus für strukturierte Ausgaben.
103 von 106 Modellen von DeepInfra unterstützen Funktionsaufrufe (Tool-Nutzung).
Ja, DeepInfra bietet 61 Reasoning-Modelle an: MiMo-V2.6-Pro, DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), GLM-5.3-Flash, Ling 3.1 Flash, Qwen3.8 2.4T A95B, DeepSeek V4.1 Flash (max), MiMo-V2.6-Flash, DeepSeek V4 Pro 0813 (max), DeepSeek V4 Flash Vision (max), DeepSeek V4 Flash 0731 (max), GLM-5.2 (max) (FP4), Qwen3.8 27B (xhigh), DeepSeek V4 Pro (max) (FP4), MiniMax-M3, GLM-5 (FP4), Kimi K2.6 (FP4), GLM-5.1 (FP4), MiMo-V2.5-Pro, Kimi K2.7 Code, Inkling Small, Hy3 (FP8), Inkling (xhigh) (FP8), DeepSeek V4 Flash (max) (FP4), Kimi K2.5, Nemotron 3 Ultra, Nemotron 3 Ultra BF16, Qwen3.5 27B (FP8), MiniMax-M2.5 (FP8), GLM-4.7 (FP4), MiMo-V2.5, DeepSeek V3.2 (FP4), Qwen3.6 27B FP8, Ling 3.0 Flash, Step 3.7 Flash, Qwen3.5 35B A3B (FP8), GLM-4.6 (FP4), Qwen3.5 397B A17B (FP8), Qwen3.6 35B A3B (FP8), Muse Glimmer (high), Gemma 4 26B A4B (FP8), Qwen3.5 122B A10B (FP4), GLM-4.7-Flash, Gemma 4 31B, Qwen3.5 4B (FP8), DeepSeek R1 0528, Nemotron 3.5 Lightning (NVFP4), Nemotron 3 Super, Qwen3 235B A22B 2507 (FP8), gpt-oss-120b (high) (Turbo), gpt-oss-120b (high), Granite 4.2 8B, gpt-oss-120b (low), Granite 4.2 3B, gpt-oss-20b (high), Gemma 4 E4B, Nemotron 3 Nano, Qwen3 30B (FP8), NVIDIA Nemotron Nano 9B V2, Qwen3 32B (FP8) und Qwen3 14B (FP8). Reasoning-Modelle nutzen ausführliches Schlussfolgern, um komplexe Probleme vor der Antwort zu bearbeiten.
Ja, 105 von 106 Modellen von DeepInfra haben offene Gewichte: MiMo-V2.6-Pro, DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), GLM-5.3-Flash, Ling 3.1 Flash, Qwen3.8 2.4T A95B, DeepSeek V4.1 Flash (max), MiMo-V2.6-Flash, DeepSeek V4 Pro 0813 (max), DeepSeek V4 Flash 0731 (max), GLM-5.2 (max) (FP4), Qwen3.8 27B (xhigh), DeepSeek V4 Pro (max) (FP4), MiniMax-M3, GLM-5 (FP4), Kimi K2.6 (FP4), GLM-5.1 (FP4), MiMo-V2.5-Pro, Kimi K2.7 Code, Inkling Small, Hy3 (FP8), Inkling (xhigh) (FP8), GLM-5.1 (non-reasoning) (FP4), DeepSeek V4 Flash (max) (FP4), Kimi K2.6 (non-reasoning) (FP4), Kimi K2.5, Nemotron 3 Ultra, Nemotron 3 Ultra BF16, Qwen3.5 27B (FP8), MiniMax-M2.5 (FP8), GLM-4.7 (FP4), MiMo-V2.5, GLM-5 (non-reasoning) (FP8), DeepSeek V3.2 (FP4), Qwen3.5 397B A17B (non-reasoning) (FP8), Qwen3.6 27B FP8, Ling 3.0 Flash, Qwen3.6 27B (non-reasoning) FP8, Step 3.7 Flash, Qwen3.5 27B (non-reasoning) FP8, Qwen3.5 35B A3B (FP8), GLM-4.6 (FP4), Qwen3.5 397B A17B (FP8), MiMo-V2.5-Pro (non-reasoning), Qwen3.6 35B A3B (FP8), Qwen3.5 122B A10B (non-reasoning) (FP4), Muse Glimmer (high), GLM-4.7 (non-reasoning) (FP4), Gemma 4 26B A4B (FP8), DeepSeek V3.2 (non-reasoning), Qwen3.5 122B A10B (FP4), Qwen3.6 35B A3B (non-reasoning) (FP8), Qwen3.5 35B A3B (non-reasoning) FP8, GLM-4.7-Flash, Gemma 4 31B, DeepSeek V3.1 Terminus (non-reasoning) (FP4), Gemma 4 31B (non-reasoning) (FP8), DeepSeek V3.1 (non-reasoning) (FP4), Gemma 4 26B A4B (non-reasoning) (FP8), Qwen3.5 4B (FP8), DeepSeek R1 0528, Nemotron 3.5 Lightning (NVFP4), Nemotron 3 Super, Qwen3 235B A22B 2507 (FP8), Qwen3 235B 2507 (FP8), Qwen3 Coder 480B (Turbo, FP4), gpt-oss-120b (high) (Turbo), gpt-oss-120b (high), Granite 4.2 8B, Qwen3.5 4B (non-reasoning) FP8, gpt-oss-120b (low), DeepSeek V3 0324 (FP4), Qwen3 Next 80B A3B, Llama 4 Maverick (FP8), Granite 4.2 3B, gpt-oss-20b (high), Gemma 4 E4B, Nemotron 3 Nano, DeepSeek V3 (Dec), Qwen2.5 72B, Llama 3.3 70B (Turbo, FP8), Qwen3 30B (FP8), Gemma 4 E4B (non-reasoning), NVIDIA Nemotron Nano 9B V2, Qwen3 32B (non-reasoning) (FP8), Qwen3 32B (FP8), Mistral Small 3.1, Mistral Small 3.2 (FP8), Llama 3.1 8B (Turbo, FP8), Llama 3.1 8B, Nemotron 3 Nano (non-reasoning), NVIDIA Nemotron Nano 9B V2 (non-reasoning), Qwen3 14B (non-reasoning) (FP8), Mistral Small 3, Qwen3 30B (non-reasoning) (FP8), Llama 3.1 70B, Llama 3.1 70B (Turbo, FP8), Llama 4 Scout, Qwen3 14B (FP8), Hermes 3 - Llama-3.1 70B, Phi-4, Llama 3.2 11B (Vision), Gemma 3 27B, Gemma 3 4B und Gemma 3 12B.
Ja, die Leistung eines Anbieters kann sich aufgrund von Infrastrukturänderungen, Lastverteilung und Aktualisierungen im Laufe der Zeit verändern. Wir benchmarken alle Anbieter fortlaufend und zeigen historische Leistungstrends in den Diagrammen „Im Zeitverlauf“ an.
Berücksichtigen Sie bei der Auswahl eines Modells von DeepInfra die Intelligenz für qualitätssensible Aufgaben, die Ausgabegeschwindigkeit für durchsatzintensive Aufgaben, die Latenz für interaktive Anwendungen mit schnellen ersten Antworten, die Preise für kostensensible Workloads sowie Funktionen wie die Größe des Kontextfensters, den JSON-Modus und die Unterstützung von Funktionsaufrufen.