Crusoe: Models Intelligence, Performance & Price
This analysis is intended to support you in choosing the best model provided by Crusoe for your use-case.
Most Intelligent
Updated
GLM-5.2 (max) (NVFP4)
DeepSeek V4 Pro 0813 (max)DeepSeek V4 Flash 0731 (max)GLM-5.1
Gemma 4 31BIntelligence index
Total 10 models
Fastest
Nemotron 3.5 Lightning (BF16)DeepSeek V4 Flash 0731 (max)
gpt-oss-120b (low)gpt-oss-120b (high)GLM-5.2 (max) (NVFP4)Output speed
Total 10 models
Lowest Price
Nemotron 3.5 Lightning (BF16)gpt-oss-120b (high)gpt-oss-120b (low)DeepSeek V4 Flash 0731 (max)Gemma 4 31BBlended price (per 1M tokens)
Total 10 models
Crusoe offers 10 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 Crusoe are GLM-5.2 (max) (NVFP4) (39), DeepSeek V4 Pro 0813 (max) (36), and DeepSeek V4 Flash 0731 (max) (35).
- For output speed, the fastest models are Nemotron 3.5 Lightning (BF16) (307 t/s), DeepSeek V4 Flash 0731 (max) (299 t/s), and gpt-oss-120b (low) (269 t/s).
- For latency, Qwen3 235B 2507 (Non-reasoning) (0.64s), Gemma 4 31B (Non-reasoning) (1.26s), and Nemotron 3.5 Lightning (BF16) (6.95s) offer the lowest time to first answer token.
- For pricing, Nemotron 3.5 Lightning (BF16) ($0.05), gpt-oss-120b (high) ($0.07), and gpt-oss-120b (low) ($0.07) offer the lowest blended prices per 1M tokens. Prices vary up to 3.3x across models.
- For context window size, DeepSeek V4 Pro 0813 (max) (1M), DeepSeek V4 Flash 0731 (max) (1M), and GLM-5.2 (max) (NVFP4) (1M) support the largest context windows on Crusoe.
- Nemotron 3.5 Lightning (BF16) offers both the fastest output and best pricing, making it attractive for throughput-sensitive and cost-conscious applications. GLM-5.2 (max) (NVFP4) leads in intelligence for tasks that require the highest quality.
Highlights
Intelligence Evaluations
Artificial Analysis Intelligence Index
Intelligence Evaluations
Agentic knowledge work, (Elo-500)/2000
Agentic real-world work tasks, (Elo-500)/2000
Agentic SaaS workflows
Agentic coding & terminal use
Coding
Reasoning & knowledge
Professional document reasoning, All-pass
Physics reasoning
Knowledge
1 - hallucination rate
Long context reasoning
Legal agentic work, criterion pass rate
Agentic business operations
Quantitative analysis on spreadsheets & documents
Instruction following
Agentic tool use
Long-horizon agentic tasks
Kubernetes incident root-cause analysis
Visual reasoning
Intelligence Index vs. Price
Context Window
Context Window
Pricing
Intelligence Index vs. Price
Performance Summary
Output Speed vs. Price
Speed
Measured by Output Speed (tokens per second)
Output Speed
Latency
Measured by Time (seconds) to First Token
Latency: Time To First Answer Token
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
Further Analysis | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
GLM-5.2 (max) (NVFP4) | 1M | Open | 39* | -- | 207 | 0.92 | 12.97 | 9.64 | |||
DeepSeek V4 Pro 0813 (max) | 1.05M | Open | 36 | $1.65 | 195 | 1.69 | 14.54 | 10.28 | |||
DeepSeek V4 Flash 0731 (max) | 1.05M | Open | 35 | $0.19 | 298 | 1.06 | 9.45 | 6.71 | |||
 Kimi K2.6 (NVFP4) | 262k | Open | 31* | -- | -- | -- | -- | -- | |||
GLM-5.1 | 131k | Open | 26 | $0.67 | 109 | 1.61 | 40.98 | 34.78 | |||
Kimi K2.6 (Non-reasoning) (NVFP4) | 262k | Open | 24* | -- | -- | -- | -- | -- | |||
Gemma 4 31B | 262k | Open | 15 | $0.10 | 26 | 1.44 | 86.53 | 66.06 | |||
Gemma 4 31B (Non-reasoning) | 262k | Open | 14* | -- | 22 | 1.21 | 24.17 | -- | |||
Nemotron 3.5 Lightning (BF16) | 262k | Open | 14 | $0.08 | 302 | 0.43 | 8.70 | 6.61 | |||
gpt-oss-120b (high) | 131k | Open | 12 | $0.04 | 256 | 0.34 | 10.11 | 7.81 | |||
 Qwen3 235B 2507 (Non-reasoning) | 262k | Open | 12* | -- | 106 | 0.67 | 5.39 | -- | |||
gpt-oss-120b (low) | 131k | Open | 10* | -- | 268 | 0.38 | 9.71 | 7.46 | |||
Key definitions
Frequently Asked Questions
Common questions about Crusoe
Crusoe offers 10 models that we track: GLM-5.2 (max) (NVFP4), DeepSeek V4 Pro 0813 (max), DeepSeek V4 Flash 0731 (max), GLM-5.1, Gemma 4 31B, Gemma 4 31B (Non-reasoning), Nemotron 3.5 Lightning (BF16), gpt-oss-120b (high), Qwen3 235B 2507 (Non-reasoning), and gpt-oss-120b (low).
The most intelligent model available on Crusoe is GLM-5.2 (max) (NVFP4) with an Intelligence Index score of 39.
The fastest model on Crusoe by output speed is Nemotron 3.5 Lightning (BF16) at 306.7 tokens per second.
The model with the lowest time to first answer token on Crusoe is Qwen3 235B 2507 (Non-reasoning) at 0.64s. Lower latency means faster initial response time.
The most affordable model on Crusoe by blended price is Nemotron 3.5 Lightning (BF16) at $0.05 per 1M tokens (7:2:1 cache hit/input/output ratio).
Prices on Crusoe vary up to 18x across models, from $0.05 per 1M tokens for Nemotron 3.5 Lightning (BF16) to $0.90 per 1M tokens for GLM-5.2 (max) (NVFP4).
Yes, Crusoe offers an OpenAI-compatible API, making it easy to switch from OpenAI or use existing OpenAI SDK integrations.
Yes, all 10 models on Crusoe support JSON mode for structured output.
9 of 10 models on Crusoe support function calling (tool use).
Yes, Crusoe offers 8 reasoning models: GLM-5.2 (max) (NVFP4), DeepSeek V4 Pro 0813 (max), DeepSeek V4 Flash 0731 (max), GLM-5.1, Gemma 4 31B, Nemotron 3.5 Lightning (BF16), gpt-oss-120b (high), and gpt-oss-120b (low). Reasoning models use extended thinking to work through complex problems before providing an answer.
Yes, all 10 models on Crusoe 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 Crusoe, 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.