Cost per task: chat reply
A short assistant turn: 400 tokens of context in, 300 tokens out. Cost = input tokens × input price + output tokens × output price. “With caching” assumes three-quarters of the input is served from the prompt cache at the provider's cache-read price.
As of 25 Oct 2026, BenchLeader ranks Granite 4.2 3B first on the Cost per task: chat reply board at $0.0000, ahead of Gemma 4 E4B ($0.0000) and Gemma 4 E2B ($0.0000).
287 models · data as of 25 Oct 2026
- 1Granite 4.2 3B$0.0000
- 2Gemma 4 E4B (thinking)$0.0000
- 3Gemma 4 E2B (thinking)$0.0000
- 4LFM2 24B A2B$0.0000
- 5Mistral Small 3$0.0000
- 6Llama 3.1 8B$0.0000
- 7MiMo-V2.6-Flash$0.0001
- 8DeepSeek V4 Flash (max)$0.0001
- 9Agnes 2.5 Pro Beta$0.0001
- 10DeepSeek V4 Flash 0420 (max)$0.0001
- 11GPT-5 nano$0.0001
- 12Quasar 438B (max)$0.0001
- 13Ling 3.0 Flash VL$0.0001
- 14MiMo-V2.5$0.0001
- 15Ling 3.0 Flash Fin$0.0001
287 of 287
| # | ||||
|---|---|---|---|---|
| 1 |  GPT-5.5 Proxhigh | $0.066 | – | 65.3 |
| 2 | GPT-5.4 Pro | $0.066 | – | 63.3 |
| 3 | GPT-5 Pro | $0.042 | – | 59.8 |
| 4 | o3-prohigh | $0.032 | – | 53.7 |
| 5 | GPT-4 | $0.030 | – | 42.4 |
| 6 | GPT 4 0613 | $0.030 | – | 39.7 |
| 7 |  Claude Opus 4.1thinking | $0.029 | $0.025 | 56.4 |
| 8 | Claude Opus 4thinking | $0.029 | $0.025 | 55.0 |
| 9 | Claude 3 Opus | $0.029 | – | 39.2 |
| 10 | o1 | $0.024 | $0.022 | 53.1 |
| 11 | Claude Fable 5.1max | $0.019 | $0.016 | 70.5 |
| 12 | GPT-6 Astramax | $0.019 | $0.016 | 70.4 |
| 13 | Claude Fable 5max | $0.019 | $0.016 | 69.3 |
| 14 | GPT-4 Turbo | $0.013 | – | 40.9 |
| 15 | GPT-5.5xhigh | $0.011 | $0.0097 | 66.1 |
| 16 | GPT-5.5 Instant | $0.011 | – | 58.0 |
| 17 | Claude Opus 5max | $0.0095 | $0.0081 | 68.8 |
| 18 | Claude Opus 4.8max | $0.0095 | $0.0081 | 63.3 |
| 19 | Claude Opus 4.7 | $0.0095 | $0.0081 | 63.2 |
| 20 | Claude Opus 4.6high | $0.0095 | $0.0081 | 62.7 |
| 21 | Claude Opus 4.5thinking | $0.0095 | $0.0081 | 59.9 |
| 22 | Claude Opus 5.5max | $0.0076 | $0.0065 | 72.6 |
| 23 | GPT-5.6 Solmax | $0.0076 | $0.0065 | 68.0 |
| 24 |  Muse Wondermax | $0.0076 | $0.0065 | 67.5 |
| 25 |  Kimi K3maxopen ↗ | $0.0057 | $0.0049 | 64.8 |
| 26 | Claude Sonnet 4.5high | $0.0057 | $0.0049 | 58.1 |
| 27 |  Grok 4 | $0.0057 | – | 58.0 |
| 28 | Claude Sonnet 4.6 | $0.0057 | $0.0049 | 57.1 |
| 29 | Claude Sonnet 4thinking | $0.0057 | $0.0049 | 53.2 |
| 30 | Grok 3 | $0.0057 | – | 51.5 |
| 31 | Claude 3.5 Sonnet | $0.0057 | – | 46.3 |
| 32 | GPT-5.4xhigh | $0.0055 | $0.0048 | 64.7 |
| 33 |  Mistral Large 24.02 | $0.0052 | – | 40.2 |
| 34 | GPT-5.3-Codexxhigh | $0.0049 | $0.0044 | 63.8 |
| 35 | GPT-5.2xhigh | $0.0049 | $0.0044 | 60.7 |
| 36 | GPT-5.2-Codexxhigh | $0.0049 | $0.0044 | 59.4 |
| 37 |  Nova Premier | $0.0048 | $0.0042 | 45.3 |
| 38 | GPT-5.6 Terramax | $0.0044 | $0.0039 | 64.4 |
| 39 |  Gemini 3.1 Pro | $0.0044 | $0.0039 | 60.5 |
| 40 | GPT-4o | $0.0040 | $0.0036 | 43.9 |
| 41 |  Command Aopen ↗ | $0.0040 | – | 41.8 |
| 42 | Gemini 4 Argonhigh | $0.0038 | $0.0032 | 70.7 |
| 43 | GPT-6.1 Solmax | $0.0038 | $0.0032 | 69.9 |
| 44 | Claude Sonnet 5.5max | $0.0038 | $0.0032 | 69.4 |
| 45 | NewKimi K3.1 Pro | $0.0038 | $0.0033 | 66.9 |
| 46 | GPT-6 Solmax | $0.0038 | $0.0033 | 66.4 |
| 47 | Claude Sonnet 5high | $0.0038 | $0.0033 | 60.1 |
| 48 | GPT-5high | $0.0035 | $0.0032 | 57.9 |
| 49 | GPT-5.1high | $0.0035 | $0.0032 | 57.7 |
| 50 | GPT-5-Codexhigh | $0.0035 | $0.0032 | 57.5 |
| 51 | GPT-5.1-Codexhigh | $0.0035 | $0.0032 | 56.3 |
| 52 | Gemini 2.5 Pro | $0.0035 | $0.0032 | 53.9 |
| 53 | Nova 2.0 Pro Previewmedium | $0.0035 | – | 50.5 |
| 54 | Gemini 3.5 Flashhigh | $0.0033 | $0.0029 | 63.1 |
| 55 |  Qwen3.7max | $0.0033 | $0.0026 | 60.6 |
| 56 | o3 | $0.0032 | $0.0028 | 59.5 |
| 57 | GPT-4.1 | $0.0032 | $0.0028 | 48.3 |
| 58 |  Jamba 1.5 Largeopen | $0.0032 | – | 42.2 |
| 59 | Jamba Large 1.6 | $0.0032 | – | 26.0 |
| 60 | Qwen3.6 Maxmax | $0.0029 | $0.0025 | 62.3 |
Cite as: BenchLeader, “Cost per task: chat reply”, data as of 25 Oct 2026, https://www.benchleader.com/leaderboards/cost-chat-turn.