LLM Speed vs Cost Calculator
Filter models by minimum speed and budget. See which models offer the best tokens-per-second for your price range, with an interactive scatter chart.
Try the Calculator↓Free · No login required · Results in seconds
Filter & Sort
Filter out models slower than this threshold
Filter out models above this price
Best for Speed
Groq — Llama 3.1 8B
750 tok/s · $0.0800/1M out
Best Value
Groq — Llama 3.1 8B
Efficiency score: 8333.3
Matching Models (26)
| Provider | Model | Speed (tok/s) | Output Cost / 1M | Efficiency Score↓ |
|---|---|---|---|---|
| Groq | Llama 3.1 8B | 750 | $0.0800 | 8333.3 |
| CHEAPESTMistral | Nemo | 150 | $0.0200 | 5000 |
| AWS Bedrock | Nova Micro | 150 | $0.1400 | 1000 |
| Gemini Flash-Lite | 200 | $0.3000 | 645.2 | |
| AWS Bedrock | Nova Lite | 120 | $0.2400 | 480 |
| Gemini Flash | 180 | $0.4000 | 439 | |
| Groq | Llama 3.3 70B | 250 | $0.7900 | 312.5 |
| xAI | Grok 2 Mini | 100 | $0.4000 | 243.9 |
| OpenAI | GPT-4o mini | 120 | $0.6000 | 196.7 |
| Mistral | Small | 120 | $0.6000 | 196.7 |
| DeepSeek | V3 | 50 | $0.2800 | 172.4 |
| Cohere | Command R | 80 | $0.6000 | 131.1 |
| AWS Bedrock | Llama 3.3 70B | 80 | $0.7200 | 109.6 |
| AWS Bedrock | Claude Haiku 3 | 100 | $1.25 | 79.4 |
| Mistral | Medium | 100 | $2.00 | 49.8 |
| AWS Bedrock | Nova Pro | 80 | $3.20 | 24.9 |
| Anthropic | Claude Haiku 4.5 | 100 | $5.00 | 20 |
| Mistral | Large | 70 | $6.00 | 11.6 |
| OpenAI | GPT-5 | 80 | $10.00 | 8 |
| OpenAI | GPT-4o | 60 | $10.00 | 6 |
| xAI | Grok 2 | 60 | $10.00 | 6 |
| Cohere | Command R+ | 60 | $10.00 | 6 |
| Gemini Pro | 70 | $12.00 | 5.8 | |
| AWS Bedrock | Mistral Large | 60 | $12.00 | 5 |
| Anthropic | Claude Sonnet 4.6 | 70 | $15.00 | 4.7 |
| AWS Bedrock | Claude Sonnet 3.5 | 70 | $15.00 | 4.7 |
* Efficiency score = tokens per second divided by output cost per 1M tokens. Higher = better value. Speeds are provider-published benchmarks as of March 2026.
Need the fastest AI at the lowest cost?
Digiqt audits your AI stack and recommends the right model mix to cut costs without sacrificing speed.
LLM Speed vs Cost — FAQ
Tokens per second (tok/s) measures how fast a model generates output. At 100 tok/s, a 200-token response takes ~2 seconds. Higher tok/s = faster responses, which matters for real-time chat and interactive applications.
Groq uses custom Language Processing Units (LPUs) specifically designed for fast sequential inference, achieving 500–800+ tok/s on 70B models. Standard GPU providers typically achieve 60–150 tok/s on comparable models.
Speed itself doesn't affect output quality — it's about hardware efficiency. The same model weights give identical output regardless of inference speed. Groq runs open-source models like Llama at much higher speeds than GPU providers.
For a good real-time chat experience, aim for 80+ tok/s. At this speed, a typical 200-token response arrives in ~2.5 seconds, which feels responsive. For streaming word-by-word, even 30–50 tok/s can feel acceptable.
Time each API request from send to final token received. Divide output token count by elapsed seconds. Most providers publish benchmark speeds, but real-world performance varies with load, model size, and prompt complexity.
Our Offices
Ahmedabad
B-714, K P Epitome, near Dav International School, Makarba, Ahmedabad, Gujarat 380051
+91 99747 29554
Mumbai
C-20, G Block, WeWork, Enam Sambhav, Bandra-Kurla Complex, Mumbai, Maharashtra 400051
+91 99747 29554
Stockholm
Bäverbäcksgränd 10 12462 Bandhagen, Stockholm, Sweden.
+46 72789 9039
Malaysia
Level 23-1, Premier Suite One Mont Kiara, No 1, Jalan Kiara, Mont Kiara, 50480 Kuala Lumpur
