DeepInfra: Models Intelligence, Performance & Price

This analysis is intended to support you in choosing the best model provided by DeepInfra for your use-case.
Most Intelligent
Intelligence Index
Total 106 models
Lowest Time per Task
Time per Intelligence Index task (minutes)
Total 106 models
Lowest Price
Blended price (per 1M tokens)
Total 106 models
DeepInfra offers 106 models, each with different intelligence, performance, and pricing characteristics. Below is a comparison of the key metrics across models.
- For intelligence, the top models on DeepInfra are Kimi K3.1 Pro (46), DeepSeek V4.1 Flash 1017 (max) (33), and GLM-5.3 (max) (32).
- For time per task, Nemotron 3.5 Lightning (NVFP4) (3.3 min), Inkling Small (3.8 min), and Nemotron 3.5 Super VL (4.0 min) finish an Intelligence Index task fastest on DeepInfra.
- For latency, Llama 3 8B (0.42s), Qwen3 30B (non-reasoning) (FP8) (0.47s), and Mistral Small 3 (0.55s) offer the lowest time to first answer token.
- For pricing, Llama 3.1 8B (Turbo, FP8) ($0.02), Llama 3.1 8B ($0.02), and Gemma 4 E4B ($0.03) offer the lowest blended prices per 1M tokens.
- For context window size, Kimi K3.1 Pro (1M), Llama 4 Maverick (FP8) (1M), and DeepSeek V4 Pro (max) (FP4) (1M) support the largest context windows on DeepInfra.
Intelligence
Artificial Analysis Intelligence Index
Evaluations
Agentic knowledge work, Elo
Knowledge
1 - hallucination rate
Agentic SaaS workflows
Agentic scientific research workflows in a terminal
Agentic software engineering in open-source repositories
Reasoning & knowledge
Professional document reasoning, All-pass
Agentic coding & terminal use
Physics reasoning
Quantitative analysis on spreadsheets & documents
Legal agentic work, Hallucination-Gated All-Pass Rate
Agentic business operations
Agentic real-world work tasks, Elo
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. Price
Pricing
Cost per Intelligence Index Task
Intelligence Index vs. Price
Speed
Average of the single-request and multi-turn agentic workloads measured
Time per Intelligence Index Task
Quality
Endpoint Accuracy Index
Latency
Average of the single-request and multi-turn agentic workloads measured
Time to First Answer Token
Reliability
Share of requests the endpoint failed, across all the single-request and multi-turn agentic workloads measured
Endpoint Error Rate
Caching
Cache Hit Rate
Cost per Task vs. Cache Hit Rate
Further Analysis | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
Kimi K3.1 Pro | 46 | $4.39 | 25.6 | 108.18 | 0.0% | 1M | Open | |||
DeepSeek V4.1 Flash 1017 (max) | 33 | $0.19 | 19.8 | 59.23 | 0.0% | 1M | Open | |||
GLM-5.3 (max) | 32 | $2.39 | 20.8 | 72.26 | 0.0% | 1M | Open | |||
MiMo-V2.6-Pro | 30 | $0.54 | 33.4 | 40.60 | 0.6% | 1M | Open | |||
GLM-5.3-Flash | 28 | $1.16 | 18.6 | 47.87 | 0.0% | 1M | Open | |||
MiMo-V2.6-Flash | 27 | $0.13 | 34.9 | 58.39 | 0.7% | 1M | Open | |||
DeepSeek V4 Pro 0813 (max) | 23 | $2.98 | 25.3 | 67.09 | 0.0% | 1M | Open | |||
Qwen3.8 27B (xhigh) | 20 | $1.54 | 37.9 | 123.64 | 0.0% | 262k | Open | |||
MiniMax-M3 | 19 | $1.11 | 39.1 | 37.71 | 1.6% | 524k | Open | |||
Nemotron 3.5 Super VL | 14 | $0.35 | 4.6 | 12.30 | 0.0% | 1M | Open | |||
Inkling Small | 13 | $0.29 | 4.1 | 24.60 | 0.9% | 524k | Open | |||
Inkling (xhigh) (FP8) | 12 | $1.26 | 5.7 | 18.91 | 0.0% | 131k | Open | |||
Nemotron 3 Ultra | 12 | $1.74 | 7.4 | 5.07 | 0.0% | 262k | Open | |||
Nemotron 3 Ultra BF16 | 12 | $4.03 | 6.6 | 4.97 | 0.0% | 262k | Open | |||
Muse Glimmer (high) | 6 | $0.13 | 5.1 | 28.79 | 1.0% | 131k | Open | |||
Nemotron 3.5 Lightning (NVFP4) | 6 | $0.31 | 3.5 | 9.14 | 0.0% | 262k | Open | |||
Granite 4.2 8B | 5 | $0.0405 | 13.7 | 57.46 | 0.0% | 131k | Open | |||
Gemma 4 31B | 4 | -- | -- | 66.90 | 0.6% | 262k | Open | |||
gpt-oss-120b (high) (Turbo) | 4 | $0.35 | 32.5 | 82.30 | 0.0% | 131k | Open | |||
gpt-oss-120b (high) | 4 | $0.0868 | 32.5 | 57.48 | 2.7% | 131k | Open | |||
Qwen3.5 122B A10B (non-reasoning) (FP4) | -- | -- | -- | 1.95 | 0.0% | 262k | Open | |||
Qwen3 Coder 480B (Turbo, FP4) | -- | -- | -- | 2.59 | 0.0% | 262k | Open | |||
Kimi K2.6 (FP4) | -- | -- | -- | 164.68 | 0.0% | 262k | Open | |||
Qwen3.5 397B A17B (FP8) | -- | -- | -- | 49.42 | 0.0% | 262k | Open | |||
Kimi K2.6 (non-reasoning) (FP4) | -- | -- | -- | 1.77 | 0.0% | 262k | Open | |||
MiniMax-M2.5 (FP8) | -- | -- | -- | 29.14 | 0.0% | 197k | Open | |||
Qwen3.5 27B (FP8) | -- | -- | -- | 31.15 | 3.7% | 262k | Open | |||
Hermes 3 - Llama-3.1 70B | -- | -- | -- | 1.31 | 0.0% | 131k | Open | |||
Gemma 3 4B | -- | -- | -- | 0.94 | 3.4% | 131k | Open | |||
Qwen3.5 122B A10B (FP4) | -- | -- | -- | 42.32 | 0.0% | 262k | Open | |||
Nemotron 3 Nano (non-reasoning) | -- | -- | -- | 0.75 | 0.0% | 262k | Open | |||
Mistral Small 3.1 | -- | -- | -- | 2.58 | 0.0% | 128k | Open | |||
Qwen3 Next 80B A3B | -- | -- | -- | 1.20 | 0.0% | 262k | Open | |||
Gemma 3 27B | -- | -- | -- | 1.95 | 0.0% | 131k | Open | |||
GLM-4.7-Flash | -- | -- | -- | 4.79 | 0.0% | 203k | Open | |||
Gemma 3 12B | -- | -- | -- | 1.54 | 0.0% | 131k | Open | |||
NVIDIA Nemotron Nano 9B V2 (non-reasoning) | -- | -- | -- | 0.76 | 0.0% | 131k | Open | |||
Qwen3.5 4B (non-reasoning) FP8 | -- | -- | -- | 1.52 | 0.0% | 262k | Open | |||
Qwen3.5 397B A17B (non-reasoning) (FP8) | -- | -- | -- | 2.01 | 0.0% | 262k | Open | |||
Qwen3 235B 2507 (FP8) | -- | -- | -- | 7.33 | 0.0% | 262k | Open | |||
gpt-oss-20b (high) | -- | -- | -- | 61.14 | 2.8% | 131k | Open | |||
Qwen3.6 27B (non-reasoning) FP8 | -- | -- | -- | 2.24 | 0.0% | 262k | Open | |||
Qwen3 30B (FP8) | -- | -- | -- | 19.24 | 0.0% | 41k | Open | |||
GLM-5 (FP4) | -- | -- | -- | 27.85 | 0.0% | 203k | Open | |||
Mistral Small 3.2 (FP8) | -- | -- | -- | 2.51 | 0.0% | 128k | Open | |||
DeepSeek V3 (Dec) | -- | -- | -- | 1.31 | 0.0% | 164k | Open | |||
Qwen3 14B (FP8) | -- | -- | -- | 15.04 | 0.0% | 33k | Open | |||
Llama 3.1 70B | -- | -- | -- | 9.60 | 2.6% | 131k | Open | |||
Llama 3.1 70B (Turbo, FP8) | -- | -- | -- | 7.99 | 0.0% | 131k | Open | |||
Qwen3.5 35B A3B (FP8) | -- | -- | -- | 38.85 | 0.0% | 262k | Open | |||
DeepSeek R1 Distill Llama 70B | -- | -- | -- | -- | -- | 131k | Open | |||
Qwen3.5 27B (non-reasoning) FP8 | -- | -- | -- | 2.17 | 0.0% | 262k | Open | |||
Gemma 4 E4B (non-reasoning) | -- | -- | -- | 1.56 | 0.0% | 131k | Open | |||
GLM-5.1 (FP4) | -- | -- | -- | 54.56 | 0.0% | 203k | Open | |||
DeepSeek V4 Pro (high) (FP4) | -- | -- | -- | -- | -- | 66k | Open | |||
Llama 3.2 11B (Vision) | -- | -- | -- | 4.25 | 0.0% | 131k | Open | |||
Nemotron 3 Super | -- | -- | -- | 10.17 | 0.0% | 262k | Open | |||
Qwen3.5 4B (FP8) | -- | -- | -- | 86.66 | 1.9% | 262k | Open | |||
Qwen3.6 35B A3B (non-reasoning) (FP8) | -- | -- | -- | 1.32 | 0.0% | 262k | Open | |||
GLM-4.6 (FP4) | -- | -- | -- | 36.63 | 2.7% | 203k | Open | |||
Step 3.7 Flash | -- | -- | -- | 45.43 | 0.0% | 256k | Open | |||
DeepSeek V3.2 (non-reasoning) | -- | -- | -- | 1.98 | 0.0% | 164k | Open | |||
Gemma 4 26B A4B (FP8) | -- | -- | -- | 30.03 | 0.0% | 262k | Open | |||
GLM-4.7 (FP4) | -- | -- | -- | 20.23 | 2.9% | 203k | Open | |||
Gemma 4 26B A4B (non-reasoning) (FP8) | -- | -- | -- | 1.38 | 0.0% | 262k | Open | |||
Llama 3.1 8B (Turbo, FP8) | -- | -- | -- | 17.34 | 2.2% | 131k | Open | |||
Llama 3.1 8B | -- | -- | -- | 16.38 | 0.0% | 131k | Open | |||
DeepSeek V3 0324 (FP4) | -- | -- | -- | 0.94 | 0.0% | 164k | Open | |||
Nemotron 3 Nano | -- | -- | -- | 0.73 | 0.0% | 262k | Open | |||
DeepSeek V4 Flash Vision (max) | -- | -- | -- | 38.27 | 0.0% | 1M | Proprietary | |||
GLM-4.7 (non-reasoning) (FP4) | -- | -- | -- | 2.44 | 0.0% | 203k | Open | |||
Llama 3 8B | -- | -- | -- | 0.41 | 0.0% | 8k | Open | |||
Qwen3.6 35B A3B (FP8) | -- | -- | -- | 34.88 | 0.8% | 262k | Open | |||
DeepSeek V4 Flash (high) (FP4) | -- | -- | -- | -- | -- | 1M | Open | |||
Qwen3.6 27B FP8 | -- | -- | -- | 38.03 | 0.0% | 262k | Open | |||
Kimi K2.7 Code | -- | -- | -- | 30.75 | 0.0% | 262k | Open | |||
Qwen3.8 2.4T A95B | -- | -- | -- | 67.59 | 0.9% | 262k | Open | |||
Llama 4 Maverick (FP8) | -- | -- | -- | 1.53 | 0.7% | 1M | Open | |||
GLM-5.1 (non-reasoning) (FP4) | -- | -- | -- | 3.37 | 0.0% | 203k | Open | |||
Llama 3.3 70B (Turbo, FP8) | -- | -- | -- | 10.67 | 0.0% | 131k | Open | |||
Kimi K2.5 | -- | -- | -- | 145.67 | 0.0% | 262k | Open | |||
Ling 3.1 Flash | -- | -- | -- | 1.08 | 0.0% | 262k | Open | |||
DeepSeek R1 0528 | -- | -- | -- | 65.44 | 2.3% | 164k | Open | |||
Llama 4 Scout | -- | -- | -- | 0.95 | 0.9% | 131k | Open | |||
Qwen3 32B (FP8) | -- | -- | -- | 25.70 | 0.0% | 41k | Open | |||
Hy3 (FP8) | -- | -- | -- | 93.14 | 2.0% | 262k | Open | |||
Qwen2.5 72B | -- | -- | -- | 1.80 | 0.0% | 33k | Open | |||
DeepSeek V4 Pro (max) (FP4) | -- | -- | -- | 19.71 | 0.0% | 1M | Open | |||
Qwen3 32B (non-reasoning) (FP8) | -- | -- | -- | 0.74 | 0.0% | 41k | Open | |||
gpt-oss-120b (low) | -- | -- | -- | 4.27 | 2.6% | 131k | Open | |||
Gemma 4 31B (non-reasoning) (FP8) | -- | -- | -- | 4.06 | 0.7% | 262k | Open | |||
DeepSeek V3.2 (FP4) | -- | -- | -- | 32.48 | 5.4% | 164k | Open | |||
Gemma 4 E4B | -- | -- | -- | 27.41 | 0.8% | 131k | Open | |||
DeepSeek V4 Flash (max) (FP4) | -- | -- | -- | 6.87 | 2.3% | 1M | Open | |||
DeepSeek V4.1 Flash (max) | -- | -- | -- | 48.25 | 0.0% | 1M | Open | |||
DeepSeek V3.1 Terminus (non-reasoning) (FP4) | -- | -- | -- | 0.97 | 0.0% | 164k | Open | |||
Ling 3.0 Flash | -- | -- | -- | 8.29 | 0.0% | 131k | Open | |||
Mistral Small 3 | -- | -- | -- | 0.56 | 0.0% | 33k | Open | |||
Qwen3 14B (non-reasoning) (FP8) | -- | -- | -- | 15.93 | 0.0% | 41k | Open | |||
Phi-4 | -- | -- | -- | 0.58 | 0.0% | 16k | Open | |||
Qwen3.5 35B A3B (non-reasoning) FP8 | -- | -- | -- | 1.12 | 0.0% | 262k | Open | |||
GLM-5 (non-reasoning) (FP8) | -- | -- | -- | 2.21 | 0.0% | 203k | Open | |||
NVIDIA Nemotron Nano 9B V2 | -- | -- | -- | 0.73 | 0.0% | 131k | Open | |||
Qwen3 30B (non-reasoning) (FP8) | -- | -- | -- | 0.47 | 0.0% | 41k | Open | |||
Qwen3 235B A22B 2507 (FP8) | -- | -- | -- | 27.27 | 0.0% | 262k | Open | |||
GLM-5.2 (max) (FP4) | -- | -- | -- | 37.72 | 0.0% | 1M | Open | |||
DeepSeek V3.1 (non-reasoning) (FP4) | -- | -- | -- | 3.68 | 0.0% | 164k | Open | |||
DeepSeek V4 Flash 0731 (max) | -- | -- | -- | 14.73 | 0.0% | 1M | Open | |||
Key definitions
Frequently Asked Questions
Common questions about DeepInfra
DeepInfra offers 106 models that we track: Kimi K3.1 Pro, DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), Inkling 2 Small, MiMo-V2.6-Pro, GLM-5.3-Flash, MiMo-V2.6-Flash, DeepSeek V4 Pro 0813 (max), Qwen3.8 27B (xhigh), MiniMax-M3, Nemotron 3.5 Super VL, Inkling Small, Inkling (xhigh) (FP8), Nemotron 3 Ultra, Nemotron 3 Ultra BF16, Muse Glimmer (high), Nemotron 3.5 Lightning (NVFP4), Granite 4.2 8B, Gemma 4 E4B, Gemma 4 31B, gpt-oss-120b (high) (Turbo), gpt-oss-120b (high), Qwen3.5 122B A10B (non-reasoning) (FP4), Qwen3 Coder 480B (Turbo, FP4), Kimi K2.6 (FP4), Qwen3.5 397B A17B (FP8), Kimi K2.6 (non-reasoning) (FP4), MiniMax-M2.5 (FP8), Qwen3.5 27B (FP8), Hermes 3 - Llama-3.1 70B, Gemma 3 4B, Qwen3.5 122B A10B (FP4), Nemotron 3 Nano (non-reasoning), Mistral Small 3.1, Qwen3 Next 80B A3B, Gemma 3 27B, GLM-4.7-Flash, Gemma 3 12B, NVIDIA Nemotron Nano 9B V2 (non-reasoning), Qwen3.5 4B (non-reasoning) FP8, Qwen3.5 397B A17B (non-reasoning) (FP8), Qwen3 235B 2507 (FP8), gpt-oss-20b (high), Qwen3.6 27B (non-reasoning) FP8, Qwen3 30B (FP8), GLM-5 (FP4), Mistral Small 3.2 (FP8), DeepSeek V3 (Dec), Qwen3 14B (FP8), Llama 3.1 70B, Llama 3.1 70B (Turbo, FP8), Qwen3.5 35B A3B (FP8), Qwen3.5 27B (non-reasoning) FP8, Gemma 4 E4B (non-reasoning), GLM-5.1 (FP4), Llama 3.2 11B (Vision), Nemotron 3 Super, Qwen3.5 4B (FP8), Qwen3.6 35B A3B (non-reasoning) (FP8), GLM-4.6 (FP4), Step 3.7 Flash, DeepSeek V3.2 (non-reasoning), Gemma 4 26B A4B (FP8), GLM-4.7 (FP4), Gemma 4 26B A4B (non-reasoning) (FP8), Llama 3.1 8B (Turbo, FP8), Llama 3.1 8B, DeepSeek V3 0324 (FP4), Nemotron 3 Nano, DeepSeek V4 Flash Vision (max), GLM-4.7 (non-reasoning) (FP4), Llama 3 8B, Qwen3.6 35B A3B (FP8), Qwen3.6 27B FP8, Kimi K2.7 Code, Qwen3.8 2.4T A95B, Llama 4 Maverick (FP8), GLM-5.1 (non-reasoning) (FP4), Llama 3.3 70B (Turbo, FP8), Kimi K2.5, Ling 3.1 Flash, DeepSeek R1 0528, Llama 4 Scout, Qwen3 32B (FP8), Hy3 (FP8), Qwen2.5 72B, DeepSeek V4 Pro (max) (FP4), Qwen3 32B (non-reasoning) (FP8), gpt-oss-120b (low), Gemma 4 31B (non-reasoning) (FP8), DeepSeek V3.2 (FP4), DeepSeek V4 Flash (max) (FP4), DeepSeek V4.1 Flash (max), DeepSeek V3.1 Terminus (non-reasoning) (FP4), Ling 3.0 Flash, Mistral Small 3, Qwen3 14B (non-reasoning) (FP8), Phi-4, Qwen3.5 35B A3B (non-reasoning) FP8, GLM-5 (non-reasoning) (FP8), NVIDIA Nemotron Nano 9B V2, Qwen3 30B (non-reasoning) (FP8), Qwen3 235B A22B 2507 (FP8), GLM-5.2 (max) (FP4), DeepSeek V3.1 (non-reasoning) (FP4), and DeepSeek V4 Flash 0731 (max).
The most intelligent model available on DeepInfra is Kimi K3.1 Pro with an Intelligence Index score of 46.
The model with the lowest time per task on DeepInfra is Nemotron 3.5 Lightning (NVFP4) at 3.3 minutes per Intelligence Index task. Time per task is the output tokens per task divided by end-to-end output speed, so lower means a task finishes sooner.
The model with the lowest time to first answer token on DeepInfra is Llama 3 8B at 0.42s. Lower latency means faster initial response time.
The most affordable model on DeepInfra by blended price is Llama 3.1 8B (Turbo, FP8) at $0.022 per 1M tokens (7:2:1 cache hit/input/output ratio).
Prices on DeepInfra vary up to 66.4x across models, from $0.022 per 1M tokens for Llama 3.1 8B (Turbo, FP8) to $1.46 per 1M tokens for Kimi K3.1 Pro.
Yes, DeepInfra offers an OpenAI-compatible API, making it easy to switch from OpenAI or use existing OpenAI SDK integrations.
Yes, all 106 models on DeepInfra support JSON mode for structured output.
104 of 106 models on DeepInfra support function calling (tool use).
Yes, DeepInfra offers 61 reasoning models: Kimi K3.1 Pro, DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), Inkling 2 Small, MiMo-V2.6-Pro, GLM-5.3-Flash, MiMo-V2.6-Flash, DeepSeek V4 Pro 0813 (max), Qwen3.8 27B (xhigh), MiniMax-M3, Nemotron 3.5 Super VL, Inkling Small, Inkling (xhigh) (FP8), Nemotron 3 Ultra, Nemotron 3 Ultra BF16, Muse Glimmer (high), Nemotron 3.5 Lightning (NVFP4), Granite 4.2 8B, Gemma 4 E4B, Gemma 4 31B, gpt-oss-120b (high) (Turbo), gpt-oss-120b (high), Kimi K2.6 (FP4), Qwen3.5 397B A17B (FP8), MiniMax-M2.5 (FP8), Qwen3.5 27B (FP8), Qwen3.5 122B A10B (FP4), GLM-4.7-Flash, gpt-oss-20b (high), Qwen3 30B (FP8), GLM-5 (FP4), Qwen3 14B (FP8), Qwen3.5 35B A3B (FP8), GLM-5.1 (FP4), Nemotron 3 Super, Qwen3.5 4B (FP8), GLM-4.6 (FP4), Step 3.7 Flash, Gemma 4 26B A4B (FP8), GLM-4.7 (FP4), Nemotron 3 Nano, DeepSeek V4 Flash Vision (max), Qwen3.6 35B A3B (FP8), Qwen3.6 27B FP8, Kimi K2.7 Code, Qwen3.8 2.4T A95B, Kimi K2.5, Ling 3.1 Flash, DeepSeek R1 0528, Qwen3 32B (FP8), Hy3 (FP8), DeepSeek V4 Pro (max) (FP4), gpt-oss-120b (low), DeepSeek V3.2 (FP4), DeepSeek V4 Flash (max) (FP4), DeepSeek V4.1 Flash (max), Ling 3.0 Flash, NVIDIA Nemotron Nano 9B V2, Qwen3 235B A22B 2507 (FP8), GLM-5.2 (max) (FP4), and DeepSeek V4 Flash 0731 (max). Reasoning models use extended thinking to work through complex problems before providing an answer.
Yes, 105 of 106 models on DeepInfra are open weight models: Kimi K3.1 Pro, DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), Inkling 2 Small, MiMo-V2.6-Pro, GLM-5.3-Flash, MiMo-V2.6-Flash, DeepSeek V4 Pro 0813 (max), Qwen3.8 27B (xhigh), MiniMax-M3, Nemotron 3.5 Super VL, Inkling Small, Inkling (xhigh) (FP8), Nemotron 3 Ultra, Nemotron 3 Ultra BF16, Muse Glimmer (high), Nemotron 3.5 Lightning (NVFP4), Granite 4.2 8B, Gemma 4 E4B, Gemma 4 31B, gpt-oss-120b (high) (Turbo), gpt-oss-120b (high), Qwen3.5 122B A10B (non-reasoning) (FP4), Qwen3 Coder 480B (Turbo, FP4), Kimi K2.6 (FP4), Qwen3.5 397B A17B (FP8), Kimi K2.6 (non-reasoning) (FP4), MiniMax-M2.5 (FP8), Qwen3.5 27B (FP8), Hermes 3 - Llama-3.1 70B, Gemma 3 4B, Qwen3.5 122B A10B (FP4), Nemotron 3 Nano (non-reasoning), Mistral Small 3.1, Qwen3 Next 80B A3B, Gemma 3 27B, GLM-4.7-Flash, Gemma 3 12B, NVIDIA Nemotron Nano 9B V2 (non-reasoning), Qwen3.5 4B (non-reasoning) FP8, Qwen3.5 397B A17B (non-reasoning) (FP8), Qwen3 235B 2507 (FP8), gpt-oss-20b (high), Qwen3.6 27B (non-reasoning) FP8, Qwen3 30B (FP8), GLM-5 (FP4), Mistral Small 3.2 (FP8), DeepSeek V3 (Dec), Qwen3 14B (FP8), Llama 3.1 70B, Llama 3.1 70B (Turbo, FP8), Qwen3.5 35B A3B (FP8), Qwen3.5 27B (non-reasoning) FP8, Gemma 4 E4B (non-reasoning), GLM-5.1 (FP4), Llama 3.2 11B (Vision), Nemotron 3 Super, Qwen3.5 4B (FP8), Qwen3.6 35B A3B (non-reasoning) (FP8), GLM-4.6 (FP4), Step 3.7 Flash, DeepSeek V3.2 (non-reasoning), Gemma 4 26B A4B (FP8), GLM-4.7 (FP4), Gemma 4 26B A4B (non-reasoning) (FP8), Llama 3.1 8B (Turbo, FP8), Llama 3.1 8B, DeepSeek V3 0324 (FP4), Nemotron 3 Nano, GLM-4.7 (non-reasoning) (FP4), Llama 3 8B, Qwen3.6 35B A3B (FP8), Qwen3.6 27B FP8, Kimi K2.7 Code, Qwen3.8 2.4T A95B, Llama 4 Maverick (FP8), GLM-5.1 (non-reasoning) (FP4), Llama 3.3 70B (Turbo, FP8), Kimi K2.5, Ling 3.1 Flash, DeepSeek R1 0528, Llama 4 Scout, Qwen3 32B (FP8), Hy3 (FP8), Qwen2.5 72B, DeepSeek V4 Pro (max) (FP4), Qwen3 32B (non-reasoning) (FP8), gpt-oss-120b (low), Gemma 4 31B (non-reasoning) (FP8), DeepSeek V3.2 (FP4), DeepSeek V4 Flash (max) (FP4), DeepSeek V4.1 Flash (max), DeepSeek V3.1 Terminus (non-reasoning) (FP4), Ling 3.0 Flash, Mistral Small 3, Qwen3 14B (non-reasoning) (FP8), Phi-4, Qwen3.5 35B A3B (non-reasoning) FP8, GLM-5 (non-reasoning) (FP8), NVIDIA Nemotron Nano 9B V2, Qwen3 30B (non-reasoning) (FP8), Qwen3 235B A22B 2507 (FP8), GLM-5.2 (max) (FP4), DeepSeek V3.1 (non-reasoning) (FP4), and DeepSeek V4 Flash 0731 (max).
Yes, provider performance can vary over time due to infrastructure changes, load balancing, and updates. We continuously benchmark all providers and display historical performance trends in the "Over Time" charts.
When choosing a model on DeepInfra, consider: intelligence (for quality-sensitive tasks), time per task (how quickly the model completes an Intelligence Index task), latency (for interactive applications requiring quick first responses), pricing (for cost-sensitive workloads), and features like context window size, JSON mode, or function calling support.