Copilot bills chat and agent usage in AI Credits since June 2026, with no default ceiling. Model your team's real usage and compare it with the same workload routed through LLM Gateway — same models, pass-through prices, prompt caching, and hard budget caps.
Business and Enterprise credit allowances vary by agreement — set yours.
Coding tools resend system prompts and repo context on nearly every request; 60% is a conservative default for agentic workloads.
$595
No spending ceiling by default — budgets are manual and off until configured.
$355
Hard budget caps per org, project, and API key — spend stops at the limit.
DevPass plans run $29–$179 per developer per month across coding agents. Your current inputs work out to about $70.98 per developer in gateway usage.
Estimate what GitHub Copilot's usage-based AI Credits cost your team each month, then compare the same workload routed through LLM Gateway in three steps.
Set your headcount, Copilot plan, and how much each developer actually uses chat and agent mode per day. Presets cover light chat through fully agentic workflows.
Pick the model mix, adjust the prompt cache hit rate, and toggle bring-your-own-keys. Every constant in the math is documented below and adjustable.
See the estimated monthly Copilot AI Credits bill next to the same workload at pass-through token prices with caching — and what a hard budget cap means for the worst case.
On June 1, 2026, GitHub Copilot moved chat, agent mode, code review, and CLI usage from flat-fee plans to metered AI Credits, where one credit is $0.01 and cost varies by model. The seat price — $10 for Pro, $39 for Pro+, $19 per user for Business, $39 per user for Enterprise — is no longer the ceiling on the bill; it's the floor. Inline completions are the only feature that stayed flat-fee.
The economics of coding assistants make this expensive fast. Every chat turn resends the conversation so far, and every agent step resends system prompts, file trees, and diffs. Token volume grows with usage squared, not linearly — which is how a team paying $50 a month under flat pricing can project $3,000 under metered billing with heavy agent use.
The same mechanics are also why routing the workload through a gateway is cheaper: all that resent context is exactly what prompt caching absorbs, billing repeated input tokens at roughly a tenth of the normal rate. Add pass-through provider pricing with no per-seat fee, and the structural gap this calculator shows emerges — before you even consider routing lighter tasks to cheaper models.
When the estimate looks right, the GitHub Copilot migration guide maps each Copilot workflow to its gateway-backed replacement, and the full comparison covers features beyond cost.
Everything you need to know about estimating and capping your team's AI coding spend.
Since June 1, 2026, GitHub Copilot bills chat, agent mode, code review, and CLI usage in AI Credits (1 credit = $0.01) on top of the seat price. Seats cost $10 (Pro), $39 (Pro+), $100 (Max), $19/user (Business), or $39/user (Enterprise). Pro includes $15 of monthly credits, Pro+ $70, and Max $200; usage beyond that bills per token with no ceiling unless you set a manual budget. Inline completions remain flat-fee.
It models a chat session as a ~5-turn conversation totaling about 30,000 input and 4,000 output tokens (history is resent every turn), and an agent task as roughly 150,000 input and 8,000 output tokens. Both Copilot AI Credits and LLM Gateway are priced from the same token volumes at the same per-million-token rates, so the comparison isolates the structural differences: seat fees and included credits versus prompt caching and a flat platform fee.
Three reasons: there's no per-seat fee for API usage, provider token rates pass through with zero markup (a flat 5% fee on credits, or 0% with your own provider keys), and prompt caching bills the repeated context that coding tools resend on every request at roughly 10% of the normal input rate. Agentic workloads resend a lot of context, so caching does most of the work.
Yes. LLM Gateway enforces budgets with hard limits per organization, project, and API key — requests stop at the cap instead of billing past it. Copilot's spending budgets exist in the billing dashboard but are off by default.
DevPass plans give each developer a flat monthly allowance ($29–$179/month) usable across coding agents like DevPass Code, Claude Code, and Cline, with roughly 3x the plan price in monthly usage value. It's the predictable-seat model Copilot used to be, but with model choice.
They're planning estimates, not invoices. Real costs depend on your models' exact rates, conversation lengths, agent context sizes, and cache hit rates — all of which vary by team. The assumptions are documented on this page and every knob is adjustable, so you can match the math to your own usage before you commit to anything.
No. Inline completions weren't moved to usage billing and remain excellent. A common setup keeps Copilot Free or a $10 Pro seat for completions and routes chat and agent workloads through LLM Gateway, where they're cached, capped, and billed at pass-through rates.
Yes — free, no signup, and everything runs in your browser. When you're ready to test real traffic, an LLM Gateway account is free to create and works with any OpenAI-compatible tool.