Back Market moved 245 developers off direct Anthropic access
What Back Market measured after 2 months on OpenRouter: a per-person budget, a provider shortlist, a longer liability chain. Full report to read online.
Jump to summary
Written by Florian Bruniaux
AI Founding Engineer at Méthode Aristote, 13 years scaling engineering teams from developer to CTO. Builds open-source developer tools, see what else I've shipped.
Full report
- Version
- 2.8.0
- Edition
- September 18, 2026
- Weight
- 3.9 MB
≈ 1 h 12 min read · full report
Estimated from the full report: 130 words/min + 30 s per code block. Excludes videos, external links and hands-on exercises.
The report opens in a new tab, with its own layout, search and glossary. Once loaded, it reads offline.
What you'll set up
- ✓ What changes when the budget sits on a person instead of a team
- ✓ The criteria that cut 70 providers down to a shortlist
- ✓ Where responsibility moves once an intermediary joins the chain
- ✓ How to repeat the measurement on your own tasks
Prerequisites
- → A team already paying for models per use
On 14/09/2026, Nicolas Martignole published a 2-month review of Back Market running on OpenRouter with 245 internal users. He gives the figures, the provider selection method and what he would do differently.
The report above takes those statements, checks them against OpenRouter’s documentation and other public accounts, and separates what is verified from what is not.
The video and the report are in French. This page summarises both in English.
TL;DR
I do not judge Back Market’s OpenRouter setup: this is a third party’s experience, and I did not run it myself.
What it shows for a team paying per use: a cap per person instead of per team, a provider shortlist driven by data policy rather than listed price, and a longer liability chain once an intermediary joins.
If you repeat the measurement, compare the cost per task actually resolved, not the price per token.
What the report establishes
Back Market replaced direct Anthropic access with an intermediary that exposes several providers behind one API. 3 decisions carry the rest:
- A budget of $200 per developer per month, reset every month, with an overrun procedure. The cap sits on the person, not on the team.
- A shortlist of about 10 providers, kept out of the 70 the catalogue offered. The criterion is each provider’s data policy, not its listed price.
- A liability chain that gets longer. The intermediary is not the model provider, and every provider on the shortlist brings its own terms.
What the report does not say
It does not compare models against each other, does not rank quality and does not recommend a provider. Amounts stay in dollars, as in the source. Any figure that no primary source confirms is marked as such.
Who it is for
For a team already paying for models per use and looking to regain control of the spend without banning the tool. The last part describes how to repeat the measurement: point Claude Code at a router, set a per-key cap, then compare the cost per task actually resolved rather than the price per token.
Go Further in the Claude Code Guide
Practical resources selected to help you take the next step.
Why this matters
The research and reasoning behind this playbook.
FinOps applied to tokens: who owns the AI bill
Cloud FinOps put a cost study first and a named owner for the number second. Three practitioner cases show that sequence applied to token bills.
Measuring your own AI usage: three meters, three different answers
Twenty days of my own Claude Code and Codex usage, measured three ways. The tools disagree by a factor of three, and dollars are not what binds.
3/6 · Mapping the token-reduction toolbox
Compare RTK, Tokenade, Headroom and prompt caching by mechanism, integration and evidence. Smaller tool output does not establish a cheaper completed task.