Cost · 2026-08-30
Cost per session
What one real AI coding session costs — median spend per model, per coding agent, per session length. Token list prices tell you the rate; this tells you the bill.
Rate is not cost
Two models with the same price per million tokens rarely cost the same to finish the same job. Reasoning length, context re-reads, tool calls, and retries all land on the invoice. OpenRouter's session-cost dataset measures the end result: the median dollars spent on a session, bucketed by how long that session ran.
In Hermes Agent, a median 10–49 turns session spans gpt-oss-20b at $0.0014 up to GPT-5.6 Sol Pro at $5.87 — a spread far wider than their list prices imply.
What a session actually costs in Hermes Agent
Median cost of one coding session by session length. 30-day window, log scale.
GPT-5.6 Sol Pro costs 4246× more per session than gpt-oss-20b in this agent.
How to read: Each row is a model. Dots run from short sessions to long ones — the further right, the more a median session costs. The number on the right is the 10–49 turns median.
- $5.87
- $4.30
- $2.18
- $1.68
- $1.67
- $1.21
- $1.19
- $1.01
- $0.981
- $0.829
- $0.797
- $0.736
- $0.703
- $0.693
Intelligence per dollar of real work
Benchmark score against the median 10–49 turns session cost in Hermes Agent.
Models on the dashed line are undominated: nothing is both cheaper and better-scoring. Bubble size is OpenRouter token share.
How to read: Cost runs left to right on a log scale, benchmark score bottom to top. Up and to the left is better value. Click a bubble for the model page.
Cost per session FAQ
- What does one AI coding session actually cost?
- In Hermes Agent, the median 10–49 turns session ranges from $0.0014 on gpt-oss-20b to $5.87 on GPT-5.6 Sol Pro, measured on 2026-08-30.
- Why measure cost per session instead of cost per million tokens?
- Token list price says nothing about how many tokens a model burns to finish a task. A cheaper model that reasons for longer, retries more, or re-reads more context can cost more per completed session than a pricier one. Session cost is the number that lands on the bill.
- Where does the session cost data come from?
- OpenRouter's session-cost dataset, which reports median spend per session grouped by coding agent, model, and session length. Tidelines re-reads it daily and joins it to usage share and benchmark scores.
- Which model gives the best intelligence per dollar?
- The frontier chart above draws the undominated set: models with no cheaper, higher-scoring alternative. Anything on that dashed line is a defensible pick; everything above it is paying for something else.
- Do longer sessions cost proportionally more?
- Not linearly. Cost grows with conversation length because context is re-sent each turn, so 50+ turn sessions typically cost several times a 10–49 turn session on the same model.