# Spend less or ship more with the models you already use.

Markdown version of https://kymaapi.com/cost-savings, for agents and crawlers. Same content as the HTML page.

- Generated: 2026-10-18T19:28:44.579Z
- HTML page: https://kymaapi.com/cost-savings
- Model catalog: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md

Move work to a cheaper model without leaving the endpoint you already call. Every figure here is measured on real Kyma sessions, not on a benchmark.

## Measured on real sessions

The calculator reads the same table.

| # | Model | Uptime 30d | Cost / session (p50) | p95 | Tokens / session | Steps / session | tok/s | Sessions |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | DeepSeek V4.1 Flash | 100.0% | $0.0005 | $0.018 | 27,378 | 2 | 36 | 988 |
| 2 | MiMo V2.5 | 100.0% | $0.0007 | $0.0007 | 4,018 | 2 | — | 3 |
| 3 | DeepSeek R1 | 100.0% | $0.0009 | $0.0027 | 462 | 12 | 56 | 236 |
| 4 | GLM 4.5 Air | 100.0% | $0.0011 | $0.0053 | 2,531 | 18 | — | 6 |
| 5 | Llama 4 Maverick | 100.0% | $0.0019 | $0.091 | 5,809 | 3 | — | 10 |
| 6 | GPT-OSS 120B | 100.0% | $0.0022 | $0.032 | 37,880 | 7 | — | 45 |
| 7 | Claude Haiku 4.5 | 100.0% | $0.012 | $0.115 | 3,555 | 4 | — | 87 |
| 8 | Sonar | 100.0% | $0.014 | $0.014 | 369 | 2 | — | 1 |
| 9 | DeepSeek V4 Flash | 100.0% | $0.014 | $0.177 | 110,531 | 10 | 22 | 803 |
| 10 | MiniMax M2.7 | 100.0% | $0.015 | $0.175 | 22,607 | 4 | — | 6 |
| 11 | Qwen 3.7 Plus | 100.0% | $0.023 | $0.235 | 16,564 | 5 | — | 11 |
| 12 | Claude Sonnet 4.6 | 100.0% | $0.024 | $0.162 | 3,207 | 2 | — | 19 |
| 13 | Claude Sonnet 5 | 100.0% | $0.044 | $0.635 | 13,131 | 3 | — | 49 |
| 14 | GLM 5.2 | 100.0% | $0.047 | $0.241 | 44,730 | 3 | 67 | 37 |
| 15 | GLM 4.7 Flash | 100.0% | $0.048 | $0.135 | 181,268 | 10 | — | 66 |
| 16 | GLM 5.3 | 100.0% | $0.092 | $1.08 | 69,781 | 7 | — | 11 |
| 17 | Grok 4.5 | 100.0% | $0.126 | $0.139 | 54,669 | 2 | — | 6 |
| 18 | Kimi K2.6 | 100.0% | $0.127 | $0.568 | 250,839 | 3 | 29 | 9 |
| 19 | Kimi K3 | 100.0% | $0.138 | $1.53 | 106,536 | 8 | — | 97 |
| 20 | MiniMax M3 | 100.0% | $0.222 | $3.01 | 447,285 | 26 | — | 5 |
| 21 | Qwen 3.6 Plus | 99.9% | $0.029 | $0.314 | 30,105 | 4 | 52 | 188 |
| 22 | GLM 5.3 Flash | 99.8% | $0.015 | $0.256 | 187,532 | 4 | — | 147 |
| 23 | Gemma 4 31B | 99.7% | $0.0018 | $0.014 | 22,747 | 2 | 8 | 127 |
| 24 | Gemini 3.8 Flash | 99.4% | $0.012 | $0.326 | 10,022 | 3 | — | 56 |
| 25 | Gemini 3.5 Flash Lite | 99.3% | $0.0028 | $0.097 | 5,033 | 3 | 48 | 273 |
| 26 | GPT-5.6 Sol | 99.2% | $0.074 | $0.715 | 30,121 | 4 | — | 235 |
| 27 | GPT-5.6 Luna | 99.0% | $0.0056 | $0.137 | 10,447 | 3 | — | 446 |
| 28 | Qwen 3 Coder | 99.0% | $0.012 | $0.103 | 28,979 | 2 | 46 | 30 |
| 29 | Qwen 3.7 Flash | 98.8% | $0.014 | $0.065 | 94,839 | 18 | — | 260 |
| 30 | Gemini 3.7 Flash | 98.4% | $0.0069 | $0.237 | 4,960 | 3 | — | 1,490 |
| 31 | Qwen 3.8 27B | 98.1% | $0.043 | $0.300 | 201,082 | 2 | — | 18 |
| 32 | Gemini 3 Flash | 97.7% | $0.018 | $1.04 | 15,136 | 9 | 39 | 62 |
| 33 | MiniMax M2.5 | 97.5% | $0.026 | $0.419 | 104,475 | 3 | — | 282 |
| 34 | GPT-5.6 Luna Pro | 97.4% | $0.032 | $0.237 | 276,473 | 3 | — | 13 |
| 35 | Qwen 3 32B | 97.0% | $0.0010 | $0.0099 | 3,653 | 4 | — | 37 |
| 36 | DeepSeek V3 | 96.3% | $0.0060 | $0.067 | 12,457 | 5 | — | 52 |
| 37 | Gemini 3.5 Flash | 96.2% | $0.0031 | $0.318 | 1,241 | 2 | — | 665 |
| 38 | DeepSeek V4 Flash Vision | 95.5% | $0.0010 | $0.221 | 5,561 | 6 | — | 76 |
| 39 | Qwen 3.8 Flash | 87.0% | $0.016 | $0.161 | 129,505 | 6 | — | 87 |
| 40 | Gemini 3.6 Flash | 87.0% | $0.205 | $5.01 | 135,377 | 13 | — | 19 |
| 41 | Llama 3.3 70B | 86.3% | $0.0019 | $0.078 | 6,115 | 7 | 14 | 69 |
| 42 | GPT-6 Astra | 4.3% | $5.66 | $5.66 | 413,437 | 19 | — | 1 |
| 43 | Muse Spark 1.1 | — | $0.0002 | $0.0002 | 57 | 2 | — | 1 |
| 44 | Grok Build | — | $0.0031 | $0.0031 | 1,317 | 6 | — | 1 |
| 45 | GLM 5.1 | — | $0.034 | $0.034 | 14,632 | 4 | — | 1 |
| 46 | Sonar Pro | — | $0.038 | $0.038 | 1,382 | 2 | — | 1 |
| 47 | GPT-5.6 Terra Pro | — | $0.060 | $0.449 | 9,877 | 4 | — | 6 |
| 48 | Qwen 3.7 Max | — | $0.084 | $0.752 | 44,735 | 2 | — | 15 |
| 49 | Gemini 3.1 Pro | — | $0.102 | $0.102 | 7,855 | 4 | — | 1 |
| 50 | Kimi K2.7 Code | — | $0.104 | $0.268 | 42,693 | 5 | — | 3 |
| 51 | Claude Opus 5 | — | $0.140 | $6.44 | 19,947 | 4 | — | 8 |
| 52 | Claude Opus 4.7 | — | $0.212 | $0.269 | 29,383 | 5 | — | 3 |
| 53 | Hermes 3 405B | — | $0.316 | $0.950 | 233,943 | 8 | — | 5 |
| 54 | Claude Fable 5.1 | — | $0.331 | $4.92 | 16,074 | 6 | — | 13 |
| 55 | GPT-5.6 Sol Pro | — | $0.424 | $0.428 | 141,372 | 2 | — | 3 |
| 56 | GPT-5.6 Terra | — | $0.596 | $0.957 | 154,789 | 3 | — | 3 |
| 57 | Grok 4.3 | — | $0.905 | $1.09 | 975,249 | 29 | — | 3 |
| 58 | Claude Fable 5 | — | $1.59 | $8.02 | 117,143 | 4 | — | 25 |
| 59 | Qwen 3.8 Max | — | $23.64 | $157.72 | 10,528,078 | 47 | — | 3 |

Measured on real Kyma sessions in the last 90 days. A session is the requests from one user and API key, cut at a 5-minute idle gap. Sessions differ per model, so this is what a session costs on that model, not the same task repriced. Uptime is 30-day, one definition, shared with /models and /status. Sessions computed 2026-10-18; prices as of 2026-10-18.