Modular: Models Intelligence, Performance & Price
This analysis is intended to support you in choosing the best model provided by Modular for your use-case.
Most Intelligent
Intelligence Index
Total 9 models
Lowest Time per Task
Time per Intelligence Index task (minutes)
Total 9 models
Lowest Price
Blended price (per 1M tokens)
Total 9 models
Modular offers 9 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 Modular are DeepSeek V4.1 Flash 1017 (max) (33), GLM-5.3 (max) (32), and MiniMax-M3.1-Flash (29).
- For time per task, Qwen3.8 27B (non-reasoning) (NVFP4) (3.4 min), DeepSeek V4.1 Flash 1017 (non-reasoning) (3.8 min), and DeepSeek V4.1 Flash 1017 (max) (7.7 min) finish an Intelligence Index task fastest on Modular. Time per task varies significantly across models, with the slowest taking 2.5x longer than the quickest.
- For latency, DeepSeek V4.1 Flash 1017 (non-reasoning) (0.62s), Gemma 4 31B (non-reasoning) (NVFP4) (0.94s), and Qwen3.8 27B (non-reasoning) (NVFP4) (1.68s) offer the lowest time to first answer token.
- For pricing, Qwen3.8 27B (xhigh) (NVFP4) ($0.00), Qwen3.8 27B (non-reasoning) (NVFP4) ($0.00), and MiniMax-M3.1-Flash ($0.10) offer the lowest blended prices per 1M tokens.
- For context window size, GLM-5.3 (max) (1M), DeepSeek V4.1 Flash 1017 (max) (1M), and MiniMax-M3.1-Flash (1M) support the largest context windows on Modular.
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 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
DeepSeek V4.1 Flash 1017 (max) | 33 | $0.84 | 8.5 | 28.75 | 0.0% | 1M | Open | |||
GLM-5.3 (max) | 32 | $5.10 | 11.4 | 29.92 | 0.0% | 1M | Open | |||
MiniMax-M3.1-Flash | 29 | $0.43 | 9.7 | 37.81 | 0.0% | 1M | Open | |||
Qwen3.8 27B (xhigh) (NVFP4) | 20 | $0.00 | 9.1 | 30.10 | 0.7% | 262k | Open | |||
MiniMax-M3 | 19 | $1.15 | 62.2 | 5.86 | 78.6% | 1M | Open | |||
DeepSeek V4.1 Flash 1017 (non-reasoning) | 19 | $0.63 | 4.4 | 0.75 | 0.0% | 1M | Open | |||
Qwen3.8 27B (non-reasoning) (NVFP4) | 10 | $0.00 | 3.4 | 1.81 | 0.0% | 262k | Open | |||
Gemma 4 31B (NVFP4) | 4 | -- | -- | 4.16 | 0.0% | 262k | Open | |||
Gemma 4 31B (non-reasoning) (NVFP4) | -- | -- | -- | 1.01 | 0.0% | 262k | Open | |||
Key definitions
Frequently Asked Questions
Common questions about Modular
Modular offers 9 models that we track: DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), MiniMax-M3.1-Flash, Qwen3.8 27B (xhigh) (NVFP4), MiniMax-M3, DeepSeek V4.1 Flash 1017 (non-reasoning), Qwen3.8 27B (non-reasoning) (NVFP4), Gemma 4 31B (NVFP4), and Gemma 4 31B (non-reasoning) (NVFP4).
The most intelligent model available on Modular is DeepSeek V4.1 Flash 1017 (max) with an Intelligence Index score of 33.
The model with the lowest time per task on Modular is Qwen3.8 27B (non-reasoning) (NVFP4) at 3.4 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 Modular is DeepSeek V4.1 Flash 1017 (non-reasoning) at 0.62s. Lower latency means faster initial response time.
The most affordable model on Modular by blended price is Qwen3.8 27B (xhigh) (NVFP4) at $0.00 per 1M tokens (7:2:1 cache hit/input/output ratio).
Prices on Modular range from $0.00 per 1M tokens for Qwen3.8 27B (xhigh) (NVFP4) to $0.90 per 1M tokens for GLM-5.3 (max).
Yes, Modular offers an OpenAI-compatible API, making it easy to switch from OpenAI or use existing OpenAI SDK integrations.
Yes, all 9 models on Modular support JSON mode for structured output.
Yes, all 9 models on Modular support function calling (tool use).
Yes, Modular offers 6 reasoning models: DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max), MiniMax-M3.1-Flash, Qwen3.8 27B (xhigh) (NVFP4), MiniMax-M3, and Gemma 4 31B (NVFP4). Reasoning models use extended thinking to work through complex problems before providing an answer.
Yes, all 9 models on Modular 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 Modular, 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.