Vamsi Penmetsa’s Post

GitHub Copilot is moving from premium requests to token-based AI Credits on 1 June. The subscription prices have not changed. What has changed is what that money buys, and it is now entirely dependent on how your developers use the tools. A quick chat and a multi-hour agentic session used to count the same. Under token metering, they will consume vastly different amounts of the same budget. This follows weeks of sign-up freezes across individual and business plans, driven by the economic reality that agentic workflows were routinely costing GitHub more per session than users paid per month. Claude Code and Cursor have already been through this transition. The consistent lesson: the billing model is not the hard part. The hard part is model governance, budget controls, and usage visibility that goes beyond a monthly total. Most organisations are not ready for AI tool costs that vary with developer behaviour rather than headcount. The ones that get governance in place before 1 June will be fine. The ones that do not will find out in July. How are you preparing for token-metered AI development tooling? #GitHubCopilot #AIGovernance #EnterpriseAI

GitHub Copilot is changing its pricing model and moving to usage-based billing from June 1, 2026. Previously, pricing was based on “premium requests”, basically the number of prompts you sent while using premium models. On the Pro plan, that was 300 per month. From June 1, this changes. Instead, every plan will include a monthly amount of “AI credits” (1 credit = $0.01). Usage will then consume credits based on the number of input, output, and cached tokens used by the model. For Business plans (Copilot Business and Enterprise), credits are pooled. Example: an enterprise with 100 Copilot Business users gets a shared pool of ~190,000 AI credits instead of 100 individual limits. Heavy users can consume more, lighter users balance it out. My personal take: Copilot will likely become noticeably more expensive with this model. And I think this is another sign that the era of flat subscription-based coding agents is ending. If you pay per use anyway, and unused credits do not roll over from month to month, then token efficiency suddenly matters much more. Especially for people using coding agents heavily or integrating them deeply into their workflow. Good news: there are already approaches to reduce token usage without losing quality. Example: https://lnkd.in/dNh2-kb2 Also: review your AGENTS.md and keep it short and precise. Disable MCP servers you don’t need now. Official sources: Press release: https://lnkd.in/deN6nPQe Pricing details: https://lnkd.in/dQ7cd97V For OpenAI models, GitHub’s listed model rates appear aligned with token-based API-style pricing: https://lnkd.in/dMbsM5d3

A week ago, I posted about LLM cost tiers, and why using the best model isn't cost effective, GitHub just proved the point. GitHub announced Copilot is moving to usage-based billing on June 1. The multiplier changes tell the whole story: https://lnkd.in/eMDzfNsE Claude Opus 4.6: 3x → 27x Claude Sonnet 4.5: 1x → 6x Claude Haiku 4.5: stays at 0.33x That's an 81x cost ratio between the most expensive and cheapest model in the same ecosystem. Defaulting to Opus for everything just became a very expensive habit. I'm living this in real time. Halfway through my billing cycle using Claude paired with Cursor's Composer 2, a Tier 2 model that handles 90% of my dev work, and I'm nowhere near my usage limits (around 20% usage of premium models). I'll likely be reducing my service payments next month. Meanwhile, developers defaulting to frontier models for routine builds, debugging, and straightforward feature work are about to get a wake-up call. GitHub said it plainly: the all-you-can-eat AI buffet is ending. The developers and companies who already treat model selection as an engineering decision won't feel it. Everyone else will. And GitHub isn't the only one, I have no doubt other tools are next. The right model for the right task isn't just a best practice anymore — it's a billing strategy. Are you rethinking your model defaults after today's news? #AIStrategy #LLMCosts #GitHubCopilot #ModelSelection #EnterpriseAI #AIEfficiency

The party is over. 🎉 GitHub Copilot just announced usage-based billing, and the new multipliers say enough: Sonnet goes from 1x to 9x, Opus from 3x to 27x. A long Copilot session that used to count as one premium request now costs what it actually costs. None of this should be a surprise. What we've had for the past two years was a subsidy. Hook first, build dependency, monetise later. GitHub's own words give it away: "align pricing with actual usage and build a sustainable business." Per-request pricing was never going to survive agentic coding. The surprise isn't that it ended, it's that it took this long. The practical consequence is straightforward: at these rates, going direct is cheaper. OpenRouter, the Anthropic API, direct OpenAI access, all of them now undercut what Copilot charges once you do the maths. And for teams with stricter data requirements, hosting models directly on Azure AI Foundry or Google Vertex gives you the same frontier models at infrastructure cost, with no platform markup on top. The IDE integration is nice, but it's not 9x nice. What it makes me wonder is what happens inside enterprises. A lot of companies are signing Copilot contracts driven by AI FOMO, leadership mandating AI adoption without a clear picture of what it costs at scale or where it actually creates value. Open-source models like DeepSeek will eventually force prices back down. But by then, most of the contracts will be signed, the vendors locked in, and the damage done.

GitHub announced that all GitHub Copilot plans are moving to usage-based billing. They framed this as a necessary move to align pricing with the shift from code assistant to agentic coding platform. The new pricing model preserves existing seat-based prices but those prices now fund a shared pool of AI credits. GitHub is effectively moving Copilot from a flat SaaS model to a hybrid of seat licensing plus cloud‑like consumption. The new structure aligns cost with actual usage of advanced, compute‑intensive features such as premium models, coding agents, and autonomous workflows. For enterprises, this introduces both flexibility and budget variability. The era of “one predictable Copilot price” is ending. Engineering leaders will need stronger visibility and governance to ensure spend aligns with value. What organizations should do now: • Understand current usage patterns and which workflows consume credits • Segment developers (baseline assistance vs agentic engineering) • Introduce usage policies and guardrails before variable spend creates surprises • Measure ROI in engineering outcomes, not just credit consumption Announcement from Github: https://lnkd.in/ewwtH_3H I didn't think I'd be right so soon with my last post.l: https://lnkd.in/ekxsBBEw #AI #GitHubCopilot #agenticcoding #softwaredevelopment

GitHub just admitted Copilot's pricing model is broken. 🚨 Starting June 1, every Copilot plan moves to usage-based billing. Fixed subscriptions are out. Token-metered "AI Credits" are in. The base prices stay the same — $10/mo for Pro, $39 for Pro+, $19/seat for Business. But now each plan gets a credit pool worth that amount. Use more, pay more. Here's the part that matters: GitHub's CPO said it directly — "the current premium request model is no longer sustainable." A quick autocomplete and a multi-hour agentic coding session were costing users the same. GitHub was eating the difference. That math stopped working. This isn't just a GitHub problem. Cursor already moved to credit pools last year. Anthropic charges Claude Code by the token. OpenAI did the same with Codex. The flat-rate era for AI dev tools is ending across the board. The real question is what this does to teams. One analyst put it well — a single developer seat now contains two completely different cost profiles. The dev nudging completions all day costs almost nothing to serve. The power user running hour-long agentic edits on a frontier model? An order of magnitude more. Same seat, wildly different bill. IDC forecasts Global 1,000 companies will underestimate AI infrastructure costs by 30% through 2027. Usage-based billing doesn't fix that — it just makes the bill more visible. 📊 If you're running Copilot on a team, start watching your token usage now. GitHub will preview the new bills in early May. What's your team's plan for managing AI tool costs as everything moves to consumption pricing? #GitHubCopilot #AITools #DevTools

GitHub Copilot is moving to usage-based billing Think of it like a mobile data plan. Right now everyone pays a flat rate regardless of whether they browse Twitter or stream 4K video all day. Starting June 1, GitHub is switching to metered billing. Heavy users pay more. Light users stop subsidizing them. The mechanism: premium requests get replaced by GitHub AI Credits, billed by token consumption. Plan prices stay the same. How you pay changes. Why now? Copilot runs multi-hour agentic coding sessions across entire repos. A quick chat question and a 2-hour autonomous session currently cost the same amount. GitHub has been absorbing that difference. That stops in June. What actually changes: • Pro: $10/month, includes $10 in monthly AI Credits • Business: $19/user/month, includes $19 in credits (temporarily $30 through August) • Enterprise: $39/user/month, includes $39 in credits (temporarily $70 through August) • Code completions stay free, no credits consumed • Credits pool across the org instead of expiring per user • Admins get hard budget caps at enterprise, cost center, and user level GitHub is launching a preview billing page in early May. If you manage Copilot licenses, check it before June 1. https://lnkd.in/dxne98Z9 #GitHubCopilot #AI #Microsoft #DeveloperTools

GitHub Copilot moving to usage-based billing is not the surprise. Microsoft was subsidizing AI inference at orders of magnitude over what Anthropic charges directly. A flat $10, $19 or $39 a month while users ran agentic sessions against Opus and Sonnet was never going to hold. The math was always on borrowed time. What's surprising to me is how little is left underneath. Opus 4.7 inside Copilot after June 1: $5 per million input tokens, $25 per million output. Identical to Anthropic's API. Sonnet 4.6: $3/$15. Identical again. They are passing provider rates through, unchanged. So the question becomes: what am I paying $10, $19, or $39 a month for, on top of metered tokens that already match the provider's own pricing? A multi-model picker is not a moat. And on agentic coding, Claude Code via Opus 4.7 lands at 87.6% on SWE-bench Verified. Copilot's published agent score sits around 56.4%. I've been writing for months that teams stitching together agent squads to redo every single SDLC process would not survive a real cost model. This is what a real cost model looks like. And it's probably just the beginning. But I'm not worried for GitHub Copilot, though. Microsoft will do what they always did: bundle the Copilot subscription to other services they sell, probably with a small discount. So, most companies will say: "we already have Copilot, we don't need anything else!"

All change for the billing in GitHub Copilot. As many people expected was coming, the period of companies such as GitHub subsidizing our AI token consumption is coming to an end. GitHub have announced that from June they will swap to token/usage based billing as opposed to their current flat rate premium request model. This means AI requests will be vary in cost based on their complexity, not just be a single fixed cost irrespective of their complexity. This means the cost of using GitHub Copilot will increase for most organisations. Every GitHub admin should take a close look at the budget and cost centre settings. For more details see https://lnkd.in/eic9rM4y

A developer opened GitHub Copilot on Monday morning and noticed his favorite AI model was gone. Replaced by a newer version that cost 7.5 times more in credits. New sign-ups to the plan had been quietly suspended. The trigger was a blog post from GitHub announcing changes to Copilot individual plans. The premium AI models that developers had been using for months were being restricted or removed. Credits that used to last all month now ran out in days. Some said this was always inevitable. AI tools have been running at a loss since launch. One commenter calculated that a handful of heavy coding sessions could burn through more compute than the entire monthly subscription cost. Another pointed out that every major AI tool launch in the last two years has been a promotional period subsidized by investor money, not a business model. The bill was always going to arrive. That argument seemed settled until the enterprise crowd pushed back. Enterprise customers pay through different channels - consolidated billing, data protection agreements, compliance packages bundled under Azure. The individual plan was always a side show. Enterprise contracts carry the business, and those prices haven't changed. A few commenters noted that most coding tasks don't need the most expensive model anyway, and developers defaulting to the biggest model available were creating their own cost problem. Most of my day is spent finding bottlenecks in businesses and optimizing them with AI. I put together a framework for non-tech founders trying to figure out where AI would actually fit in their business - without having to follow every new release. Comment "AI Framework" or DM me and I'll send it over - please make sure I can DM you.