DeepInfra: Models Intelligence, Performance & Price

DeepInfra
DeepInfra

Analysis of DeepInfra's models across key metrics including quality, price, output speed, latency, context window & more. This analysis is intended to support you in choosing the best model provided by DeepInfra for your use-case.

Most Intelligent

Updated
#1
GLM-5.2 (max) (FP8)
GLM-5.2 (max) (FP8)
51
#2
DeepSeek V4 Pro (Max) (FP4)
DeepSeek V4 Pro (Max) (FP4)
44
#3
Kimi K2.6 (FP4)
Kimi K2.6 (FP4)
43
#4
MiMo-V2.5-Pro
MiMo-V2.5-Pro
42
#5
Kimi K2.7 Code
Kimi K2.7 Code
42

Intelligence index

Total 90 models

Fastest

#1
NVIDIA Nemotron Nano 12B v2 VL (FP8)
NVIDIA Nemotron Nano 12B v2 VL (FP8)
281 t/s
#2
NVIDIA Nemotron Nano 12B v2 VL (FP8)
NVIDIA Nemotron Nano 12B v2 VL (FP8)
276 t/s
#3
Llama 3.1 Nemotron 70B
Llama 3.1 Nemotron 70B
253 t/s
#4
gpt-oss-120b (high) (Turbo)
gpt-oss-120b (high) (Turbo)
232 t/s
#5
MiMo-V2.5
MiMo-V2.5
169 t/s

Output speed

Total 90 models

Lowest Price

#1
Qwen3.5 0.8B (FP8)
Qwen3.5 0.8B (FP8)
$0.01
#2
Qwen3.5 0.8B FP8
Qwen3.5 0.8B FP8
$0.01
#3
Llama 3.1 8B (Turbo, FP8)
Llama 3.1 8B (Turbo, FP8)
$0.02
#4
Llama 3.1 8B
Llama 3.1 8B
$0.02
#5
Qwen3.5 2B (FP8)
Qwen3.5 2B (FP8)
$0.03

Blended price (per 1M tokens)

Total 90 models

Indicates a reasoning model

DeepInfra offers 90 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 GLM-5.2 (max) (FP8) (51), DeepSeek V4 Pro (Max) (FP4) (44), Kimi K2.6 (FP4) (43).
  • For output speed, the fastest models are NVIDIA Nemotron Nano 12B v2 VL (FP8) (281 t/s), NVIDIA Nemotron Nano 12B v2 VL (FP8) (276 t/s), Llama 3.1 Nemotron 70B (253 t/s). Speed varies significantly across models, with a 66% difference between the fastest and slowest.
  • For latency, NVIDIA Nemotron 3 Nano (0.38s), NVIDIA Nemotron Nano 12B v2 VL (FP8) (0.43s), NVIDIA Nemotron Nano 12B v2 VL (FP8) (0.44s) offer the lowest time to first token.
  • For pricing, Qwen3.5 0.8B (FP8) ($0.01), Qwen3.5 0.8B FP8 ($0.01), Llama 3.1 8B (Turbo, FP8) ($0.02) offer the lowest blended prices per 1M tokens.
  • For context window size, GLM-5.2 (max) (FP8) (1m), DeepSeek V4 Flash (Max) FP4 (1m), DeepSeek V4 Flash (High) (FP4) (1m) support the largest context windows on DeepInfra.

Highlights

Updated

Intelligence

Artificial Analysis Intelligence Index · Higher is better

Speed

Output tokens per second · Higher is better

Price

USD per 1M tokens (blended) · Lower is better

Intelligence Evaluations

Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index · Higher is better
Estimate (independent evaluation forthcoming)

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Intelligence Evaluations

Intelligence evaluations measured independently by Artificial Analysis · Higher is better
GDPval-AA v2Updated

Agentic real-world work tasks, (Elo-500)/2000

𝜏³-BankingNew

Agentic tool use

Terminal-Bench v2.1New

Agentic coding & terminal use

SciCode

Coding

Humanity's Last Exam

Reasoning & knowledge

GPQA Diamond

Scientific reasoning

CritPt

Physics reasoning

AA-Omniscience Accuracy

Knowledge

AA-Omniscience Non-Hallucination Rate

1 - hallucination rate

AA-LCR

Long context reasoning

IFBench

Instruction following

APEX-Agents-AA

Long-horizon agentic tasks

ITBench-AA

Kubernetes incident root-cause analysis

MMMU-Pro

Visual reasoning

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Intelligence vs. Price

Blended at 7:2:1 (cache-input-output) · Artificial Analysis Intelligence Index · USD per 1M tokens (blended)
Most attractive quadrant
Reasoning models are indicated by a lightbulb icon.

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

The blended cache price shown here uses cache hit price only. Other caching costs differ by provider:

  • Anthropic: charges a separate cache write fee, with different rates for 5-minute and 1-hour TTLs (1-hour TTL is more expensive).
  • Google (Vertex/Gemini): charges a per-hour cache storage fee in addition to cache hit pricing. Some providers also use tiered pricing for prompts above 200K tokens.
  • OpenAI, DeepSeek, others: typically charge only cache hit pricing with no write or storage fee.

See Prompt Caching for the full breakdown.

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Context Window

Context Window

Context window: tokens limit · Higher is better

Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).

JSON Mode & Function Calling

Function (Tool) Calling & JSON Mode

ModelsFunction callingJSON Mode
GLM-5.2 (max) (FP8), DeepInfra logoGLM-5.2 (max) (FP8), DeepInfra
DeepSeek V4 Pro (Max) (FP4), DeepInfra logoDeepSeek V4 Pro (Max) (FP4), DeepInfra
Kimi K2.6 (FP4), DeepInfra logoKimi K2.6 (FP4), DeepInfra
MiMo-V2.5-Pro, DeepInfra logoMiMo-V2.5-Pro, DeepInfra
Kimi K2.7 Code, DeepInfra logoKimi K2.7 Code, DeepInfra
DeepSeek V4 Pro (High) (FP4), DeepInfra logoDeepSeek V4 Pro (High) (FP4), DeepInfra
DeepSeek V4 Flash (Max) FP4, DeepInfra logoDeepSeek V4 Flash (Max) FP4, DeepInfra
GLM-5.1 (FP4), DeepInfra logoGLM-5.1 (FP4), DeepInfra
MiMo-V2.5, DeepInfra logoMiMo-V2.5, DeepInfra
GLM-5 FP8, DeepInfra logoGLM-5 FP8, DeepInfra
Kimi K2.5, DeepInfra logoKimi K2.5, DeepInfra
Nemotron 3 Ultra, DeepInfra logoNemotron 3 Ultra, DeepInfra

Indicates whether the provider supports function calling in their API. Function calling is also known as 'Tool Calling'.

Indicates whether the provider supports JSON mode in their API. When JSON mode is enabled, the models will always return a valid JSON object.

Pricing

Intelligence vs. Price

Blended at 7:2:1 (cache-input-output) · Artificial Analysis Intelligence Index · USD per 1M tokens (blended)
Most attractive quadrant
Reasoning models are indicated by a lightbulb icon.

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

The blended cache price shown here uses cache hit price only. Other caching costs differ by provider:

  • Anthropic: charges a separate cache write fee, with different rates for 5-minute and 1-hour TTLs (1-hour TTL is more expensive).
  • Google (Vertex/Gemini): charges a per-hour cache storage fee in addition to cache hit pricing. Some providers also use tiered pricing for prompts above 200K tokens.
  • OpenAI, DeepSeek, others: typically charge only cache hit pricing with no write or storage fee.

See Prompt Caching for the full breakdown.

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Performance Summary

Output Speed vs. Price

Output speed: output tokens per second · USD per 1M tokens (blended) · 10,000 input tokens
Most attractive quadrant
DeepInfra
DeepInfra (FP4)
DeepInfra (FP8)
DeepInfra FP4
DeepInfra FP8

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

Speed

Measured by Output Speed (tokens per second)

Output Speed

Output tokens per second · Higher is better · 10,000 input tokens

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Latency

Measured by Time (seconds) to First Token

Time to First Answer Token

Seconds to first token received · Lower is better

Time to first answer token received, in seconds, after API request sent. For reasoning models, this includes the 'thinking' time of the model before providing an answer. For models which do not support streaming, this represents time to receive the completion.

Seconds to receive a 500 token response. Key components:

  • Input time: Time to receive the first response token
  • Thinking time (only for reasoning models): Time reasoning models spend outputting tokens to reason prior to providing an answer. Amount of tokens based on the average reasoning tokens across a diverse set of 60 prompts (methodology details).
  • Answer time: Time to generate 500 output tokens, based on output speed

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

End-to-End Response Time

Seconds to output 500 tokens, calculated based on time to first token, 'thinking' time for reasoning models, and output speed

End-to-End Response Time vs. Price

End-to-end response time: end-to-end seconds to output 500 tokens · USD per 1M tokens (blended)
Most attractive quadrant
DeepInfra
DeepInfra (FP4)
DeepInfra (FP8)
DeepInfra FP4
DeepInfra FP8

Seconds to receive a 500 token response. Key components:

  • Input time: Time to receive the first response token
  • Thinking time (only for reasoning models): Time reasoning models spend outputting tokens to reason prior to providing an answer. Amount of tokens based on the average reasoning tokens across a diverse set of 60 prompts (methodology details).
  • Answer time: Time to generate 500 output tokens, based on output speed

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Key definitions

Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Time to first token received, in seconds, after API request sent. For reasoning models which share reasoning tokens, this will be the first reasoning token. For models which do not support streaming, this represents time to receive the completion.

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

Price per token generated by the model (received from the API), represented as USD per million Tokens.

Price per token included in the request/message sent to the API, represented as USD per million Tokens.

Metrics are 'live' and are based on the past 72 hours of measurements, measurements are taken 8 times a day for single requests and 2 times per day for parallel requests.

Frequently Asked Questions

Common questions about DeepInfra

DeepInfra offers 90 models that we track: GLM-5.2 (max), DeepSeek V4 Pro (Max), Kimi K2.6, MiMo-V2.5-Pro, Kimi K2.7 Code, DeepSeek V4 Pro (High), DeepSeek V4 Flash (Max), GLM-5.1, MiMo-V2.5, GLM-5, Kimi K2.5, Nemotron 3 Ultra, DeepSeek V4 Flash (High), Qwen3.6 27B, GLM-5.1, Kimi K2.6, GLM-4.7, Qwen3.5 27B, Qwen3.5 397B A17B, MiniMax-M2.5, GLM-5, Qwen3.5 122B A10B, Qwen3.5 397B A17B, Qwen3.6 35B A3B, Gemma 4 31B, Qwen3.5 27B, Qwen3.6 27B, Qwen3.5 35B A3B, Qwen3.5 122B A10B, MiMo-V2.5-Pro, GLM-4.7, Gemma 4 26B A4B, GLM-4.6, Gemma 4 31B, DeepSeek V3.2, Qwen3.6 35B A3B, gpt-oss-120b (high), gpt-oss-120b (high), Qwen3.5 35B A3B, GLM-4.7-Flash, Qwen3 235B A22B 2507, DeepSeek V3.1 Terminus, DeepSeek V3.1, Gemma 4 26B A4B, Qwen3.5 4B, DeepSeek R1 0528, Qwen3 235B 2507, Qwen3 Coder 480B, NVIDIA Nemotron 3 Nano, Qwen3.5 4B, DeepSeek V3 0324, gpt-oss-20B (high), Llama 4 Maverick, Qwen3 Next 80B A3B, Llama Nemotron Super 49B v1.5, Qwen3 32B, DeepSeek V3 (Dec), Qwen3.5 2B, Qwen3 14B, Llama 4 Scout, DeepSeek R1 Distill Llama 70B, Qwen2.5 72B, Qwen3 30B, Mistral Small 3.2, NVIDIA Nemotron Nano 12B v2 VL, NVIDIA Nemotron Nano 9B V2, Qwen3.5 2B, Llama Nemotron Super 49B v1.5, Qwen3 32B, Llama 3.3 70B, Mistral Small 3.1, Llama 3.1 Nemotron 70B, NVIDIA Nemotron 3 Nano, NVIDIA Nemotron Nano 9B V2, Qwen3 14B, Mistral Small 3, Qwen3 30B, Llama 3.1 70B, Llama 3.1 70B, Llama 3.1 8B, Llama 3.1 8B, Hermes 3 - Llama-3.1 70B, Qwen3.5 0.8B, Phi-4, Gemma 3 27B, NVIDIA Nemotron Nano 12B v2 VL, Qwen3.5 0.8B, Gemma 3 12B, Llama 3.2 11B (Vision), and Gemma 3 4B.

The most intelligent model available on DeepInfra is GLM-5.2 (max) with an Intelligence Index score of 51.

The fastest model on DeepInfra by output speed is NVIDIA Nemotron Nano 12B v2 VL at 281.4 tokens per second.

The model with the lowest time to first token on DeepInfra is NVIDIA Nemotron 3 Nano at 0.38s. Lower latency means faster initial response time.

The most affordable model on DeepInfra by blended price is Qwen3.5 0.8B at $0.01 per 1M tokens (7:2:1 cache hit/input/output ratio).

Prices on DeepInfra vary up to 86x across models, from $0.01 per 1M tokens for Qwen3.5 0.8B to $1.20 per 1M tokens for Llama 3.1 Nemotron 70B.

Yes, DeepInfra offers an OpenAI-compatible API, making it easy to switch from OpenAI or use existing OpenAI SDK integrations.

Yes, all 90 models on DeepInfra support JSON mode for structured output.

86 of 90 models on DeepInfra support function calling (tool use).

Yes, DeepInfra offers 41 reasoning models: GLM-5.2 (max), DeepSeek V4 Pro (Max), Kimi K2.6, MiMo-V2.5-Pro, Kimi K2.7 Code, DeepSeek V4 Pro (High), DeepSeek V4 Flash (Max), GLM-5.1, MiMo-V2.5, GLM-5, Kimi K2.5, Nemotron 3 Ultra, DeepSeek V4 Flash (High), Qwen3.6 27B, GLM-4.7, Qwen3.5 27B, Qwen3.5 397B A17B, MiniMax-M2.5, Qwen3.5 122B A10B, Qwen3.6 35B A3B, Gemma 4 31B, Qwen3.5 35B A3B, Gemma 4 26B A4B, GLM-4.6, gpt-oss-120b (high), gpt-oss-120b (high), GLM-4.7-Flash, Qwen3 235B A22B 2507, Qwen3.5 4B, DeepSeek R1 0528, NVIDIA Nemotron 3 Nano, gpt-oss-20B (high), Llama Nemotron Super 49B v1.5, Qwen3 32B, Qwen3.5 2B, Qwen3 14B, DeepSeek R1 Distill Llama 70B, Qwen3 30B, NVIDIA Nemotron Nano 12B v2 VL, NVIDIA Nemotron Nano 9B V2, and Qwen3.5 0.8B. Reasoning models use extended thinking to work through complex problems before providing an answer.

Yes, all 90 models on DeepInfra are open weight models.

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), output speed (for throughput-intensive tasks), 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.