MEASURED IN PRODUCTION · REFRESHED WEEKLY · WINDOW ENDS 2026-10-11
WHAT A SESSION ACTUALLY COSTS
Benchmarks measure what a task costs on a model's best day. This is the other number: the median dollars a real session burns in 4 agent harnesses, by model and by how long the session runs. Half of all sessions cost more than the figure shown. A 50+ turn session runs from $0.018 (Codex, deepseek-v4-flash) to $21.79 (Claude Code, claude-fable-5.1).
Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-12. Licensed CC BY 4.0. OpenRouter is a sample, not the market: traffic routed through one aggregator, skewed towards price-sensitive and open-model usage. Direct enterprise traffic to OpenAI, Anthropic, Google or Azure does not appear here. How we read third-party data →
CLAUDE CODE
Median USD per session · 45 models with enough sessions to publish
| Model | 1 turn | 2–9 turns | 10–49 turns | 50+ turns |
|---|---|---|---|---|
| anthropic/claude-fable-5.1 | $0.303 | $1.02 | $3.60 | $21.79 |
| openai/gpt-6-astra | $0.214 | $0.521 | $2.68 | $14.28 |
| anthropic/claude-4.8-opus | $0.167 | $0.443 | $1.89 | $11.42 |
| anthropic/claude-opus-5 | $0.173 | $0.410 | $2.04 | $10.98 |
| anthropic/claude-4.6-opus | $0.047 | $0.352 | $1.50 | $10.75 |
| anthropic/claude-opus-5.5 | $0.134 | $0.304 | $1.38 | $8.95 |
| qwen/qwen3.8-max | $0.050 | $0.138 | $1.03 | $5.95 |
| moonshotai/kimi-k3 | $0.048 | $0.129 | $0.884 | $5.30 |
| anthropic/claude-5-fable | $0.055 | $0.851 | $4.97 | — |
| anthropic/claude-sonnet-5.5 | $0.069 | $0.158 | $0.622 | $4.74 |
| anthropic/claude-4.6-sonnet | $0.094 | $0.190 | $0.788 | $4.62 |
| openai/gpt-5.6-sol | $0.050 | $0.146 | $0.683 | $4.01 |
| anthropic/claude-sonnet-5 | $0.093 | $0.161 | $0.657 | $3.67 |
| z-ai/glm-5.3 | $0.015 | $0.057 | $0.462 | $3.32 |
| google/gemini-3.8-flash | $0.0038 | $0.023 | $0.299 | $3.07 |
| openai/gpt-6.1-sol | $0.043 | $0.103 | $0.402 | $2.64 |
| z-ai/glm-5.2 | $0.013 | $0.061 | $0.339 | $2.62 |
| deepseek/deepseek-v4-pro-20260813 | $0.011 | $0.036 | $0.277 | $1.67 |
| minimax/minimax-m3 | — | $0.020 | $0.154 | $1.40 |
| anthropic/claude-4.7-opus | — | $0.465 | $1.38 | — |
| deepseek/deepseek-v4-pro-20260423 | $0.0074 | $0.024 | $0.183 | $1.38 |
| x-ai/grok-4.6 | — | $0.148 | $1.33 | — |
| anthropic/claude-4.5-haiku | $0.012 | $0.047 | $0.258 | $1.27 |
| tencent/hy4-preview | — | — | $0.255 | $1.21 |
| x-ai/grok-4.7 | — | $0.108 | $1.09 | — |
| qwen/qwen3.8-27b | — | $0.015 | $0.154 | $1.03 |
| anthropic/claude-haiku-5.5 | $0.0023 | $0.0076 | $0.055 | $0.818 |
| anthropic/claude-4.5-sonnet | $0.100 | $0.161 | $0.690 | — |
| xiaomi/mimo-v2.6-pro | $0.0080 | $0.020 | $0.107 | $0.564 |
| qwen/qwen3.8-flash | $0.0036 | $0.0079 | $0.060 | $0.474 |
| deepseek/deepseek-v4.1-flash | $0.0037 | $0.0096 | $0.056 | $0.467 |
| z-ai/glm-5.3-flash | $0.0024 | $0.0084 | $0.058 | $0.454 |
| openai/gpt-5.6-luna | $0.0048 | $0.011 | $0.047 | $0.353 |
| meta/muse-spark-1.3-contributor | $0.0020 | $0.0076 | $0.046 | $0.295 |
| moonshotai/kimi-k2.7-code | — | $0.034 | $0.291 | — |
| openai/gpt-6-luna | $0.0021 | $0.0049 | $0.026 | $0.272 |
| deepseek/deepseek-v4-flash-20260423 | $0.0009 | $0.0038 | $0.029 | $0.245 |
| xiaomi/mimo-v2.6-flash | $0.0024 | $0.0074 | $0.036 | $0.189 |
| deepseek/deepseek-v4-flash-20260731 | $0.0009 | $0.0043 | $0.031 | $0.185 |
| openai/gpt-6-luna-pro | — | $0.014 | $0.079 | — |
| openai/gpt-6-sol | — | $0.077 | — | — |
| moonshotai/kimi-k2.6 | — | $0.032 | — | — |
| deepseek/deepseek-chat-v3 | $0.0029 | — | — | — |
| openai/gpt-oss-120b | — | $0.0022 | — | — |
| qwen/qwen3.7-flash | — | $0.0018 | — | — |
CODEX
Median USD per session · 23 models with enough sessions to publish
| Model | 1 turn | 2–9 turns | 10–49 turns | 50+ turns |
|---|---|---|---|---|
| openai/gpt-6-astra | $0.127 | $0.596 | $2.79 | $18.61 |
| openai/gpt-5.6-sol | $0.024 | $0.121 | $0.580 | $3.46 |
| openai/gpt-6-sol | $0.025 | $0.110 | $0.578 | $3.27 |
| openai/gpt-5.6-terra | $0.024 | $0.132 | $0.488 | $2.67 |
| openai/gpt-6.1-sol | $0.028 | $0.106 | $0.476 | $2.47 |
| google/gemini-3.8-flash | $0.0092 | $0.041 | $0.311 | $1.41 |
| z-ai/glm-5.3 | $0.0054 | $0.028 | $0.232 | $0.794 |
| moonshotai/kimi-k3 | $0.014 | — | $0.562 | — |
| anthropic/claude-opus-5.5 | $0.057 | $0.559 | — | — |
| openai/gpt-5.3-codex | $0.016 | $0.063 | $0.409 | — |
| openai/gpt-5.6-luna | $0.0025 | $0.0093 | $0.042 | $0.249 |
| openai/gpt-6-luna | $0.0015 | $0.0051 | $0.030 | $0.234 |
| deepseek/deepseek-v4.1-flash | $0.0009 | $0.0067 | $0.039 | $0.108 |
| meta/muse-spark-1.3-contributor | $0.0012 | $0.0040 | $0.014 | $0.074 |
| z-ai/glm-5.3-flash | $0.0007 | $0.0048 | $0.032 | $0.064 |
| x-ai/grok-4.7 | $0.033 | — | — | — |
| deepseek/deepseek-v4-flash-20260731 | $0.0004 | $0.0028 | — | $0.018 |
| openai/gpt-5.4-mini | $0.015 | — | — | — |
| deepseek/deepseek-v4-pro-20260423 | $0.0046 | — | — | — |
| deepseek/deepseek-v4-pro-20260813 | $0.0035 | — | — | — |
| openai/gpt-6-luna-pro | $0.0024 | — | — | — |
| xiaomi/mimo-v2.6-pro | $0.0011 | — | — | — |
| deepseek/deepseek-v4-flash-20260423 | $0.0004 | — | — | — |
HERMES AGENT
Median USD per session · 120 models with enough sessions to publish
| Model | 1 turn | 2–9 turns | 10–49 turns | 50+ turns |
|---|---|---|---|---|
| openai/gpt-6-astra | $0.011 | $0.474 | $2.93 | $18.69 |
| anthropic/claude-fable-5.1 | $0.0070 | $0.504 | $3.23 | $16.17 |
| anthropic/claude-opus-5 | $0.017 | $0.278 | $2.06 | $11.15 |
| anthropic/claude-4.8-opus | $0.0064 | $0.267 | $1.53 | $9.70 |
| anthropic/claude-opus-5.5 | $0.021 | $0.202 | $1.32 | $8.69 |
| openai/gpt-6-astra-pro | $0.034 | $0.926 | $8.49 | — |
| openai/gpt-5.5 | $0.027 | $0.224 | $1.30 | $7.05 |
| x-ai/grok-4.6 | $0.010 | $0.129 | $0.994 | $5.92 |
| anthropic/claude-4.6-opus | $0.0025 | $0.114 | $0.869 | $5.29 |
| x-ai/grok-4.7 | $0.0100 | $0.091 | $0.992 | $5.12 |
| anthropic/claude-sonnet-5.5 | $0.018 | $0.086 | $0.547 | $4.27 |
| anthropic/claude-4.6-sonnet | $0.047 | $0.110 | $0.715 | $3.97 |
| anthropic/claude-5-fable | $0.013 | $0.717 | $3.81 | — |
| openai/gpt-5.6-sol | $0.021 | $0.094 | $0.561 | $3.77 |
| anthropic/claude-sonnet-5 | $0.022 | $0.096 | $0.628 | $3.70 |
| moonshotai/kimi-k3 | $0.018 | $0.087 | $0.583 | $3.48 |
| meta/muse-spark-1.3 | $0.016 | $0.073 | $0.571 | $3.30 |
| qwen/qwen3.8-max | $0.0040 | $0.063 | $0.476 | $3.22 |
| openai/gpt-6.1-sol | $0.021 | $0.067 | $0.417 | $3.13 |
| openai/gpt-5.6-terra | $0.021 | $0.093 | $0.474 | $2.93 |
| openai/gpt-5.6-sol-pro | — | $0.295 | $2.92 | — |
| openai/gpt-6-sol | $0.0098 | $0.062 | $0.415 | $2.81 |
| anthropic/claude-4-sonnet | $0.020 | $0.110 | $0.579 | $2.79 |
| z-ai/glm-5.3 | $0.0041 | $0.036 | $0.294 | $2.11 |
| openai/gpt-5.6-luna-pro | $0.0053 | $0.025 | $0.168 | $1.95 |
| google/gemini-3.5-flash | $0.0098 | $0.067 | $0.473 | $1.93 |
| z-ai/glm-5.2 | $0.0004 | $0.027 | $0.248 | $1.60 |
| deepseek/deepseek-v4-pro-20260813 | $0.0032 | $0.026 | $0.215 | $1.47 |
| anthropic/claude-4.5-haiku | $0.014 | $0.034 | $0.188 | $1.33 |
| deepseek/deepseek-v4-pro-20260423 | $0.0037 | $0.022 | $0.180 | $1.29 |
| moonshotai/kimi-k2.7-code | — | $0.023 | $0.244 | $1.27 |
| google/gemini-3.8-flash | $0.0060 | $0.028 | $0.208 | $1.22 |
| moonshotai/kimi-k2.6 | $0.0059 | $0.026 | $0.187 | $1.21 |
| openai/gpt-6-luna-pro | $0.0022 | $0.0081 | $0.080 | $1.16 |
| google/gemini-3.7-flash | $0.0070 | $0.032 | $0.195 | $1.13 |
| openai/gpt-5.4-mini | $0.0004 | $0.024 | $0.133 | $0.868 |
| z-ai/glm-5.3-flashx | $0.0022 | $0.015 | $0.111 | $0.867 |
| tencent/hy4-preview | $0.0067 | $0.026 | $0.167 | $0.818 |
| qwen/qwen3.8-27b | $0.0010 | $0.0095 | $0.103 | $0.804 |
| anthropic/claude-haiku-5.5 | $0.0014 | $0.0043 | $0.031 | $0.674 |
| anthropic/claude-4.5-sonnet | $0.030 | $0.100 | $0.630 | — |
| x-ai/grok-4.5 | $0.034 | $0.106 | $0.629 | — |
| google/gemini-3.1-pro-preview | $0.019 | $0.082 | $0.618 | — |
| deepseek/deepseek-v3.2 | $0.0026 | $0.0096 | $0.099 | $0.577 |
| openai/gpt-5.4 | — | $0.112 | $0.557 | — |
| google/gemini-3-flash-preview | $0.0017 | $0.017 | $0.081 | $0.557 |
| minimax/minimax-m3 | $0.0023 | $0.011 | $0.098 | $0.557 |
| google/gemini-2.5-flash | $0.0031 | $0.012 | $0.069 | $0.528 |
| xiaomi/mimo-v2.6-pro | $0.0021 | $0.011 | $0.089 | $0.445 |
| xiaomi/mimo-v2.5-pro | $0.0058 | $0.015 | $0.070 | $0.416 |
| google/gemini-2.5-pro | $0.025 | $0.068 | $0.403 | — |
| deepseek/deepseek-v4-flash-vision-exp | $0.0016 | $0.0082 | $0.067 | $0.400 |
| qwen/qwen3.7-max | $0.020 | $0.071 | $0.396 | — |
| nvidia/nemotron-3-super-120b-a12b | $0.0014 | $0.0063 | $0.069 | $0.373 |
| deepseek/deepseek-v4.1-flash | $0.0008 | $0.0063 | $0.052 | $0.322 |
| tencent/hy3 | $0.0006 | $0.0068 | $0.051 | $0.293 |
| openai/gpt-5.6-luna | $0.0021 | $0.0086 | $0.041 | $0.290 |
| z-ai/glm-5.3-flash | $0.0008 | $0.0058 | $0.041 | $0.287 |
| nvidia/nemotron-3-ultra-550b-a55b | $0.0043 | $0.015 | $0.254 | — |
| meta/muse-spark-1.3-contributor | $0.0008 | $0.0051 | $0.038 | $0.242 |
| qwen/qwen3.8-flash | $0.0003 | $0.0038 | $0.032 | $0.212 |
| openai/gpt-6-luna | $0.0010 | $0.0031 | $0.022 | $0.208 |
| qwen/qwen3.7-flash | $0.0003 | $0.0011 | $0.026 | $0.189 |
| qwen/qwen3.6-plus-04-02 | — | $0.020 | $0.188 | — |
| xiaomi/mimo-v2.6-flash | $0.0011 | $0.0036 | $0.030 | $0.178 |
| x-ai/grok-4.3 | $0.020 | $0.042 | $0.173 | — |
| deepseek/deepseek-v4-flash-20260731 | $0.0003 | $0.0030 | $0.024 | $0.167 |
| google/gemini-3.6-flash | $0.0016 | $0.017 | $0.158 | — |
| moonshotai/kimi-k2.5-0127 | $0.0058 | $0.017 | $0.156 | — |
| openai/gpt-6.1-sol-pro | — | $0.154 | — | — |
| deepseek/deepseek-v4-flash-20260423 | $0.0006 | $0.0035 | $0.025 | $0.153 |
| deepseek/deepseek-chat-v3.1 | — | $0.011 | $0.117 | — |
| deepseek/deepseek-chat-v3 | $0.0048 | $0.012 | $0.105 | — |
| xiaomi/mimo-v2.5 | $0.0006 | $0.0041 | $0.021 | $0.098 |
| google/gemini-2.5-flash-lite | $0.0002 | $0.0030 | $0.017 | $0.094 |
| google/gemma-4-31b-it | $0.0014 | $0.0066 | $0.088 | — |
| qwen/qwen3.7-plus | $0.0047 | $0.017 | $0.086 | — |
| minimax/minimax-m2.7 | $0.0041 | $0.012 | $0.076 | — |
| qwen/qwen3-coder-next | — | $0.0050 | $0.071 | — |
| inclusionai/ling-3.0-flash-vl | — | $0.0009 | $0.0096 | $0.064 |
| inclusionai/ling-3.0-flash | $0.0001 | $0.0009 | $0.0092 | $0.061 |
| qwen/qwen3.6-35b-a3b | — | $0.0070 | $0.058 | — |
| openai/gpt-5-mini | $0.0036 | $0.0095 | $0.055 | — |
| z-ai/glm-5.1 | — | $0.054 | — | — |
| stepfun/step-3.7-flash | — | $0.0076 | $0.050 | — |
| openai/gpt-4.1-mini | $0.0017 | $0.0079 | $0.049 | — |
| upstage/solar-pro4 | — | $0.0041 | $0.048 | — |
| openai/gpt-4o | $0.0080 | $0.043 | — | — |
| x-ai/grok-4.20 | — | $0.043 | — | — |
| deepseek/deepseek-r1 | — | $0.042 | — | — |
| google/gemini-3.5-flash-lite | $0.0020 | $0.0088 | $0.040 | — |
| openai/gpt-4o-mini | $0.0009 | $0.0039 | $0.035 | — |
| google/gemini-3.1-flash-lite | $0.0010 | $0.0058 | $0.029 | — |
| meta-llama/llama-3.3-70b-instruct | $0.0018 | $0.0060 | $0.028 | — |
| mistralai/mistral-large-4-0 | — | $0.021 | — | — |
| openai/gpt-oss-120b | $0.0004 | $0.0017 | $0.020 | — |
| inception/mercury-2.5 | $0.0004 | $0.0018 | $0.020 | — |
| z-ai/glm-4.6 | — | $0.011 | — | — |
| minimax/minimax-m2.5 | — | $0.010 | — | — |
| openai/gpt-oss-20b | $0.0001 | $0.0008 | $0.0098 | — |
| qwen/qwen3-coder-480b-a35b-07-25 | — | $0.0098 | — | — |
| deepseek/deepseek-chat-v3-0324 | — | $0.0081 | — | — |
| google/gemini-3.1-flash-lite-preview | — | $0.0077 | — | — |
| openai/gpt-5.4-nano | $0.0001 | $0.0056 | — | — |
| qwen/qwen3-coder-30b-a3b-instruct | — | $0.0045 | — | — |
| qwen/qwen3-235b-a22b-07-25 | $0.0008 | $0.0043 | — | — |
| meta-llama/llama-3.1-8b-instruct | — | $0.0038 | — | — |
| z-ai/glm-4.7-flash | — | $0.0034 | — | — |
| qwen/qwen3.5-flash | — | $0.0033 | — | — |
| mistralai/mistral-small-2603 | — | $0.0032 | — | — |
| qwen/qwen3-vl-32b-instruct | — | $0.0028 | — | — |
| openai/gpt-5-nano | $0.0007 | $0.0023 | — | — |
| google/gemma-4-26b-a4b-it | — | $0.0020 | — | — |
| nvidia/nemotron-3.5-lightning | — | $0.0020 | — | — |
| mistralai/mistral-small-3.2-24b-instruct-2506 | — | $0.0020 | — | — |
| poolside/laguna-s-2.1 | — | $0.0020 | — | — |
| openai/gpt-4.1-nano | $0.0001 | $0.0018 | — | — |
| qwen/qwen3-30b-a3b-instruct-2507 | — | $0.0018 | — | — |
| venice/uncensored | $0.0001 | $0.0006 | — | — |
| mistralai/mistral-nemo | $0.0000 | $0.0005 | — | — |
KILO CODE
Median USD per session · 26 models with enough sessions to publish
| Model | 1 turn | 2–9 turns | 10–49 turns | 50+ turns |
|---|---|---|---|---|
| anthropic/claude-opus-5.5 | — | $0.386 | $1.61 | $7.95 |
| anthropic/claude-sonnet-5.5 | — | $0.200 | $0.851 | $5.33 |
| moonshotai/kimi-k3 | — | $0.198 | $0.875 | $4.88 |
| anthropic/claude-sonnet-5 | — | $0.202 | $0.782 | $4.59 |
| openai/gpt-6-astra | — | $0.822 | $2.98 | — |
| z-ai/glm-5.3 | $0.025 | $0.084 | $0.451 | $2.62 |
| anthropic/claude-opus-5 | — | — | $2.28 | — |
| google/gemini-3.8-flash | $0.0008 | $0.034 | $0.300 | $1.70 |
| deepseek/deepseek-v4-pro-20260813 | — | $0.047 | $0.228 | $1.68 |
| deepseek/deepseek-v4-pro-20260423 | — | $0.053 | $0.212 | $1.51 |
| openai/gpt-5.6-sol | — | $0.323 | $0.786 | — |
| openai/gpt-6.1-sol | — | $0.153 | $0.503 | — |
| xiaomi/mimo-v2.6-pro | — | $0.036 | $0.110 | $0.493 |
| z-ai/glm-5.3-flash | $0.0012 | $0.010 | $0.052 | $0.408 |
| openai/gpt-5.6-luna | $0.0004 | $0.016 | $0.055 | $0.370 |
| deepseek/deepseek-v4.1-flash | $0.0013 | $0.013 | $0.052 | $0.307 |
| meta/muse-spark-1.3-contributor | — | $0.0079 | $0.034 | $0.304 |
| z-ai/glm-5.2 | — | $0.058 | $0.283 | — |
| openai/gpt-6-luna | $0.0002 | $0.0081 | $0.034 | $0.259 |
| deepseek/deepseek-v4-flash-20260731 | $0.0002 | $0.0053 | $0.032 | $0.242 |
| deepseek/deepseek-v4-flash-20260423 | $0.0002 | $0.0055 | $0.027 | $0.224 |
| xiaomi/mimo-v2.6-flash | — | $0.011 | $0.036 | $0.184 |
| minimax/minimax-m3 | — | $0.018 | $0.077 | — |
| qwen/qwen3.8-flash | — | — | $0.041 | — |
| anthropic/claude-4.5-haiku | $0.0019 | — | — | — |
| google/gemini-nano-banana-2.1 | $0.0016 | — | — | — |
Read this carefully
- OpenRouter is a sample, not the market: traffic routed through one aggregator, skewed towards price-sensitive and open-model usage. Direct enterprise traffic to OpenAI, Anthropic, Google or Azure does not appear here.
- Each cell is the MEDIAN spend of sampled sessions for one harness × model × turn range; half the sessions cost more. Sessions are never pooled across harnesses.
- Refreshed weekly by OpenRouter over a rolling sample window (see source.window_days).
- A cell exists only where OpenRouter had enough sessions to publish a median without exposing anyone. A missing cell means "not enough data", never "free".
- Session cost is harness × model × your prompts. The same model in two harnesses can differ several-fold, which is the point: the harness decides how many tokens you burn. See the Burn Index →
Dataset v1 · 30-day sample window ending 2026-10-11 · generated 2026-10-14 · 648 cells