The end of cheap, unlimited AI coding
„The era of fixed monthly fees for generous AI coding has ended: token-based billing, hidden budget cuts, and corporate spending caps are coming.”
Counter-narrative
Restrictions are not one-way. Anthropic extended the weekly usage bonus for Claude Code until August 31. On October 30, OpenAI removed the Codex limit from the Pro 200 plan and doubled Work and Codex limits for Pro and Business plans. A Hungarian-language report states that Databricks reduced AI coding costs by 70% using open-source models. Costs can thus be lowered by vendor decisions and model selection.
Carriers
GitHub Copilot pricing decisions, Anthropic Claude Code-limitek, OpenAI Codex-limitek, Apple, Microsoft, Meta internal budgets, The New Stack, DEV Community, Practical Tools, Gartner
Analysis
Providers are moving one after another from fixed fees to credits and tokens: Copilot, Codex and Cursor are all taking this path. Meanwhile, they are narrowing generous limits or making them temporary. Extreme cases show what agent-based work can cost. One Codex run cost $19,300, and one Claude Code skill consumed 200,000 tokens before a single question. The shift places the cost risk on the user. Communication, such as the Claude Code limit announced as an increase but reported to be a decrease, breeds distrust. On the corporate side, cost ceilings are appearing: Devin introduced hierarchical budgets, while Apple, Microsoft and Meta have tightened internal limits. The opposite trend is real too: OpenAI doubled its limits, and Databricks significantly cut its costs with open models. The weak point is comparing total cost against return, because the stories contain no return data.
What's at stake: Whether the unit cost of AI coding is rising faster than returns, thereby curbing corporate adoption.
Narrative sub-threads
Quiet tightening of subscription limits
In fixed-price packages, providers are tightening usage limits. Bonuses are expiring. The Claude Code change, communicated by Anthropic as a 25% increase, is reported as a 17% cut. New 5-hour, weekly, and shared ChatGPT–Codex caps are emerging. The process is not one-way: OpenAI reversed an announced Pro restriction and doubled limits.
Here, the price remains while the quantity within the package decreases. The evidence comes from limit changes and user experiences, not price lists or corporate decisions.
→ lasting · evidence: strongShift to credit- and token-based billing
Leading coding tools are switching from monthly fees to credit, token, and consumption-based billing, so the fee follows actual usage.
This strand shows a structural rewrite of the pricing model (credits, token price, agent billing), mainly based on the GitHub/Microsoft and OpenAI Codex price lists.
↗ rising · evidence: moderateCorporate AI cost caps and budgets
Companies are increasingly addressing the runaway costs of agents. Apple has set daily spending limits, and per-developer caps appear in the examples. According to reports, Meta and Microsoft are curbing internal Claude usage. On the vendor side, Devin is implementing hierarchical cost budgets.
The actor here is the user company and the CTO, not the provider. The evidence consists of corporate spending ranges, reports on internal restrictions, and ROI doubts.
→ lasting · evidence: moderateContext and language token costs
Hidden costs arise in token-based billing. Skills and MCP servers already consume large amounts of tokens before meaningful work begins (200,000 and 18,000 tokens respectively). Non-English languages incur additional costs. API message threads do not reduce the token bill even if the entire conversation is sent only once.
This branch shows the technical sources of consumption based on individual cases and developer measurements, not pricing or corporate decisions.
↗ rising · evidence: moderateShift toward cheaper models and routes
The market responds to rising prices with cheaper open-source and Chinese models, proxies, model routing, and provider switching. Savings claims typically range between 30–90%.
The mechanism here is circumvention and substitution, not restriction. The evidence is typically vendor savings claims and individual switches; Replit's switch is disputed by insiders.
Sub-thread stories
Quiet tightening of subscription limits
Anthropic extends Claude Code weekly usage boost through August 31tech leadBefore the bonus expires, assess the team's actual usage so that limits returning to normal do not stall work.startup founderThe higher allowance lasts only until the end of August, so it should not be treated as permanent when planning costs.Anthropic raises Claude Code and Opus API rate limits, citing SpaceX Colossus 1 dealsenior developerCheck the new limits that apply to your own plan, as the increase is not the same for everyone.tech leadReassess how much load parallel agents and long sessions can take.CTOTake the provider's capacity expansion into account when choosing vendors.OpenAI restores 5-hour Codex and Work limits for ChatGPT Plus userssenior developerPace Codex use within the limits, or move to a higher plan if your workflow requires it.tech leadFactor the limits into team planning and check whether the more expensive plans pay off.startup founderDo not base cost plans on the discounted limits.Claude Code weekly limit rises 25%, but reports say real capacity falls 17%senior developerBefore 14 September, measure your actual weekly consumption and reschedule long agent runs.tech leadRedo the team's Claude Code quota planning and designate a fallback tool for when the limit is hit.CTOThe cost model is changing, so recalculate licence costs and the split between tools.startup founderIf development relies on a single tool's limit, a tighter quota can slow delivery.Anthropic promised 20x more usage, but developers hit a weekly capsenior developerBecause of the weekly cap, long agent jobs must be scheduled, or a fallback model kept ready.tech leadPlans with a visible cost ceiling are more predictable for the team.Claude Code weekly limit drops: the '25% increase' was in effect a 17% cutjunior developerThe weekly allowance may run out faster, so long agent runs should be planned more deliberately.senior developerScheduling of large agent tasks needs to be adjusted to the tighter weekly allowance.tech leadMeasure the team's actual Claude Code consumption again before planning a sprint on the old allowance.CTOChanges to provider limits are a cost and predictability risk, so it is worth keeping a fallback tool.Claude Code cloud sessions open to all users; overtime mode draws on weekly limitsenior developerPlan your weekly usage allowance, because extra work is drawn from it.OpenAI reverses ChatGPT Pro Codex cut and doubles usage limitssenior developerOn the Pro subscription the Codex allowance is not reduced and in fact increases; on the Plus plan it is unchanged.CTOWhen comparing subscription and API costs, account for the doubled limits.OpenAI to pool ChatGPT and Codex usage allowance from Decembersenior developerCodex and chat draw on one allowance, so schedule heavy agent tasks deliberately.tech leadRecalculate the team's subscription allowance, because Codex use will crowd out other ChatGPT use.
Shift to credit- and token-based billing
Copilot: credit complaints, token budget and new Lite and Balanced tiersCTOCopilot costs need to be recalculated based on the credit model and the new tiers.tech leadIntroduce a per-developer token or credit budget and a policy on model choice.senior developerRestrict the use of expensive models to tasks that genuinely need them.startup founderLocal LLMs or API-based billing can be considered as alternatives.GitHub Copilot's monthly usage hits record high after new pricing modelCTOUnder the new pricing, track teams' actual consumption, because costs can rise with usage.tech leadPermissions and usage need to be monitored to avoid unexpected bills.Copilot harness reaches general availability in JetBrains; Copilot Studio billing starts September 1CTOPrepare a cost estimate for Copilot Studio harness agents before billing starts on 1 September.tech leadAssess which agents are built on the Copilot harness and how much they are used.senior developerThe harness is now available in JetBrains IDEs and can be tried out.AI coding tools move to usage-based pricing: Copilot credits and token ratesCTOMonthly AI cost now depends on usage, so introduce budgets and reporting.tech leadGive the team a rule on when a premium model may be used and when a cheaper one is enough.senior developerSave premium models for hard tasks and use a cheaper model for routine work.startup founderDeveloper tooling costs now scale with usage, so choose between Copilot, Cursor and Claude Code on cost.Copilot team AI credits cut by up to 44% at an unchanged priceCTOThe effective unit price has risen, so the Copilot budget needs to be recalculated.tech leadTrack credit consumption and schedule expensive agent tasks more sparingly.Copilot budget increase requests now generally available, adding another AI consumption unit to tracktech leadThe team also needs to follow the budget increase request process and unit-based consumption.CTOManaging AI costs is more complicated with new consumption units, so budget limits should be reviewed.GitHub Copilot adds cost and quality controls, and Code Review gains agent capabilitiestech leadSet a budget and an approval process for extra credits, and try out agentic Code Review.CTOCredit-based billing requires cost oversight, and the new controls can help with that.senior developerAgentic review can complement manual review, but it does not replace it.Codex moves to token-based billing from April, with credits and plan limitsCTOCost depends on usage, so plan a budget and set up alerts.tech leadMeasure the team's token consumption before choosing a plan.Microsoft redesigns Copilot with Home, Code and Autopilot; main features are usage-basedCTOAllow for cost swings from usage-based pricing when planning licensing.tech leadAssess whether Code and Autopilot replace your existing Copilot workflows.Agent pull requests overtook human ones on GitHub in August as Copilot moves to usage-based pricingtech leadGiven the share of agent PRs, review capacity and PR rules need to be redesigned.CTOWith usage-based pricing, Copilot costs depend on load, so a spending cap should be set.senior developerMore agent PRs need to be reviewed, so review priorities should be set down in writing.
Corporate AI cost caps and budgets
AI coding agents overrun budgets: how Replit, Kilo Code and Databricks control costsCTOWithout per-developer spending caps and model routing, the cost of agent use cannot be controlled.tech leadSimpler tasks can be steered to a cheaper or open model.Companies spend millions on AI use but can barely prove its impactCTOToken consumption can be a misleading performance metric; impact should be measured by delivered results.tech leadDon't rank usage; measure lead time and defect rates.OpenAI Codex developer says two parallel agents is the limit after a $19,300 runtech leadSet cost caps and alerts for agent runs.CTOWith parallel agents, costs can escalate quickly.Meta and Microsoft scale back internal use of Anthropic's ClaudeCTOLarge companies' internal tool choices may shift, so vendor dependence should be managed.tech leadTeams should prepare their workflows for the possibility of switching providers.Gartner: billing pitfalls are the main problem when choosing enterprise AI toolsCTOBefore signing a contract, examine usage-based billing and the cost impact of limits.AI coding tools cost developers from $150 to over $3,000 a monthCTOIntroduce per-developer budgets and usage reporting, because costs differ by orders of magnitude.tech leadUse cheaper or local models for simpler tasks.startup founderBudget monthly AI tool costs as a separate line next to payroll.Amazon tells its 'Renaissance Developer' not to chase token usagetech leadMeasure the team by business outcomes rather than AI-usage metrics.senior developerBe able to justify when using AI is not worthwhile.CTOToken-consumption targets can distort behaviour, so review the metrics.Coding agents get spending limits as Cognition adds hierarchical budgets to Devintech leadSet budgets at team and task level, and define what happens to an interrupted agent task.Apple, Microsoft and Meta tighten internal AI usage limitsCTOEven large companies are keeping token costs in check, so plan how you will measure return on investment.senior developerExpect agent usage to be capped, and expect efficient prompting and context management to matter.
Context and language token costs
Claude Code skill consumed 200,000 tokens before answering a single questionsenior developerMeasure the token consumption of installed skills and trim context that is unnecessarily large.tech leadThe team's shared set of skills should go through an approval and cost-tracking process.Anthropic tokens cost more on Vercel, and non-English prompts may add hidden costsstartup founderCheck the actual token price of using a model through a platform, and the cost of non-English prompts.tech leadMeasure token cost by platform and by language before budgeting.MCP becomes the standard, but token consumption and testing infrastructure remain gapssenior developerThe token cost of MCP servers has to be measured, because large tool descriptions consume the context.tech leadTesting and cost-measurement practices need to be set up for MCP servers.'Hungarian AI tax': translation proxy reduces errors in Hungarian-language agent taskssenior developerWhen working with Hungarian-language instructions, it is worth comparing the results and cost of English prompts.junior developerWhere possible, give the agent the task in English.Anthropic adds message threads to its API, but the token bill stays the samesenior developerIf you build your own agent on the Claude API, threads may speed up response times, but do not expect lower costs.startup founderToken costs have to be optimized another way, because threads help with latency, not with the bill.
Shift toward cheaper models and routes
NVIDIA: open models and local autonomous software development with FactoryCTOFor sensitive code, consider agents running locally on open models instead of cloud-based ones.Local proxy lets Claude Code run on free modelssenior developerWorth trying for cost reduction, but code and prompts pass through a third-party proxy.Cognition says its SWE-2 coding agent matches rivals at a fraction of the costCTOThe vendor's benchmark and cost claims should be checked on your own tasks before deciding on a model switch.tech leadA pilot should compare quality and running costs against the current agent.Together Link runs open models inside Claude Code; Together AI says coding costs could fall by over 50%startup founderOpen models can cut agent costs significantly, but quality has to be measured.tech leadTools can be made model-agnostic, which reduces vendor lock-in.GitHub Copilot routes requests across models with HydraFusion, cutting estimated cost by 67% on cheapest pathCTORouting between models is one way to cut AI tooling costs and is worth building into the cost model.tech leadThe 67% figure is an estimate, so measure it against your own workload before it goes into the cost plan.Replit CEO says OpenAI is now its default model instead of Anthropicstartup founderDon't tie yourself to a single model provider; build an abstraction layer that lets you switch on cost.CTOReview your model-provider dependency and cost structure.Microsoft's new tool can cut AI coding costs by up to 31%CTONew tools can cut AI coding costs, but 31% is only an upper bound, so teams need to measure their own savings.Revolut switches from Claude Code to Chinese open-weight model, cutting costs by 90%CTOAssess how much an open-weight alternative could save compared with expensive closed models, and what it means for data protection and compliance.tech leadRun a comparison pilot on your own codebase to measure cost and quality.
Google search interest
Hungary
| Topic and search term | Weekly | Average |
|---|---|---|
| The end of cheap, unlimited AI codingAI subscription price | 0.3 | |
| Quiet tightening of subscription limitsclaude limit | 0.2 | |
| Shift to credit- and token-based billingAI token price | 0 | |
| Corporate AI cost caps and budgetsAI cost | 0 | |
| Context and language token coststoken | 24 | |
| Shift toward cheaper models and routescheap AI model | 0 |
World
| Topic and search term | Weekly | Average |
|---|---|---|
| The end of cheap, unlimited AI codingai coding pricing | 20.6 | |
| Quiet tightening of subscription limitsclaude usage limits | 43.3 | |
| Shift to credit- and token-based billingusage based pricing | 9.4 | |
| Corporate AI cost caps and budgetsai spending budget | 9.5 | |
| Context and language token costsllm token cost | 0.8 | |
| Shift toward cheaper models and routescheap ai models | 10.1 |
Google Trends relative search index over the period studied, weekly average. A narrative and its sub-threads appear in a single comparison: 100 is the group's most-searched day, and the figures can only be compared within the group. The search term was chosen by machine analysis; it measures how much the term is searched, not how widespread the narrative is.
Who is affected and how
- CTO
- Implement per-developer and team-level budgets, alerts, and monthly cost/PR metrics. Gartner specifically warns against billing traps in enterprise selection.
- tech lead
- Parallel agents and large-context skills are the main cost drivers. One Codex developer considers two parallel agents a practical limit, and one skill consumed 200,000 tokens before a single query. Therefore, skills must be audited for token consumption.
- senior developer
- It is worth choosing a cheaper model suited to the task for routine work rather than using the default top-tier model for everything.
- startup founder
- If your product runs on a lab API or subscription, pricing and limits can change unilaterally. OpenAI, for example, tightened restrictions within a week and then withdrew them. Unit economics must be scaled across multiple providers.
Hungary
No or minimal domestic trace during the reviewed period: no [HU]-marked source among the reviewed stories.
Weekly trend
Turning points
Claude Code weekly limit rises 25%, but reports say real capacity falls 17%Anthropic announced a 25% weekly limit increase starting September 14. Multiple reports indicate this is actually a 17% decrease compared to previous limits, placing the discrepancy between limit communication and the actual limit at the center of attention.Agent pull requests overtook human ones on GitHub in August as Copilot moves to usage-based pricingReports say agent PRs surpassed human ones in August, and Microsoft is introducing consumption-based pricing for Copilot. This reinforces the industry trend toward usage-based pricing.Apple, Microsoft and Meta tighten internal AI usage limitsApple, Microsoft, and Meta are also tightening their own internal AI use. Apple is limiting engineers' daily AI spending, Microsoft has reportedly cut some of its quotas by 90%, and at Meta the number of Claude Code users has halved. Cost limits have thus appeared even within the largest tech companies.
Related narratives in the News magazine
Supporting stories
AI coding agents overrun budgets: how Replit, Kilo Code and Databricks control costsCTOWithout per-developer spending caps and model routing, the cost of agent use cannot be controlled.tech leadSimpler tasks can be steered to a cheaper or open model.Copilot: credit complaints, token budget and new Lite and Balanced tiersCTOCopilot costs need to be recalculated based on the credit model and the new tiers.tech leadIntroduce a per-developer token or credit budget and a policy on model choice.senior developerRestrict the use of expensive models to tasks that genuinely need them.startup founderLocal LLMs or API-based billing can be considered as alternatives.Claude Code skill consumed 200,000 tokens before answering a single questionsenior developerMeasure the token consumption of installed skills and trim context that is unnecessarily large.tech leadThe team's shared set of skills should go through an approval and cost-tracking process.Anthropic extends Claude Code weekly usage boost through August 31tech leadBefore the bonus expires, assess the team's actual usage so that limits returning to normal do not stall work.startup founderThe higher allowance lasts only until the end of August, so it should not be treated as permanent when planning costs.OpenAI restores 5-hour Codex and Work limits for ChatGPT Plus userssenior developerPace Codex use within the limits, or move to a higher plan if your workflow requires it.tech leadFactor the limits into team planning and check whether the more expensive plans pay off.startup founderDo not base cost plans on the discounted limits.Claude Code weekly limit rises 25%, but reports say real capacity falls 17%senior developerBefore 14 September, measure your actual weekly consumption and reschedule long agent runs.tech leadRedo the team's Claude Code quota planning and designate a fallback tool for when the limit is hit.CTOThe cost model is changing, so recalculate licence costs and the split between tools.startup founderIf development relies on a single tool's limit, a tighter quota can slow delivery.AI coding tools move to usage-based pricing: Copilot credits and token ratesCTOMonthly AI cost now depends on usage, so introduce budgets and reporting.tech leadGive the team a rule on when a premium model may be used and when a cheaper one is enough.senior developerSave premium models for hard tasks and use a cheaper model for routine work.startup founderDeveloper tooling costs now scale with usage, so choose between Copilot, Cursor and Claude Code on cost.Copilot team AI credits cut by up to 44% at an unchanged priceCTOThe effective unit price has risen, so the Copilot budget needs to be recalculated.tech leadTrack credit consumption and schedule expensive agent tasks more sparingly.Anthropic promised 20x more usage, but developers hit a weekly capsenior developerBecause of the weekly cap, long agent jobs must be scheduled, or a fallback model kept ready.tech leadPlans with a visible cost ceiling are more predictable for the team.Claude Code weekly limit drops: the '25% increase' was in effect a 17% cutjunior developerThe weekly allowance may run out faster, so long agent runs should be planned more deliberately.senior developerScheduling of large agent tasks needs to be adjusted to the tighter weekly allowance.tech leadMeasure the team's actual Claude Code consumption again before planning a sprint on the old allowance.CTOChanges to provider limits are a cost and predictability risk, so it is worth keeping a fallback tool.Copilot budget increase requests now generally available, adding another AI consumption unit to tracktech leadThe team also needs to follow the budget increase request process and unit-based consumption.CTOManaging AI costs is more complicated with new consumption units, so budget limits should be reviewed.GitLab joins other providers in imposing rate limits on AI coding agentstech leadPlatform limits must be factored in when scheduling agent traffic.CTOBecause of platform restrictions, AI agent capacity should be planned vendor by vendor.OpenAI Codex developer says two parallel agents is the limit after a $19,300 runtech leadSet cost caps and alerts for agent runs.CTOWith parallel agents, costs can escalate quickly.Codex moves to token-based billing from April, with credits and plan limitsCTOCost depends on usage, so plan a budget and set up alerts.tech leadMeasure the team's token consumption before choosing a plan.Agent pull requests overtook human ones on GitHub in August as Copilot moves to usage-based pricingtech leadGiven the share of agent PRs, review capacity and PR rules need to be redesigned.CTOWith usage-based pricing, Copilot costs depend on load, so a spending cap should be set.senior developerMore agent PRs need to be reviewed, so review priorities should be set down in writing.Gartner: billing pitfalls are the main problem when choosing enterprise AI toolsCTOBefore signing a contract, examine usage-based billing and the cost impact of limits.AI coding tools cost developers from $150 to over $3,000 a monthCTOIntroduce per-developer budgets and usage reporting, because costs differ by orders of magnitude.tech leadUse cheaper or local models for simpler tasks.startup founderBudget monthly AI tool costs as a separate line next to payroll.OpenAI reverses ChatGPT Pro Codex cut and doubles usage limitssenior developerOn the Pro subscription the Codex allowance is not reduced and in fact increases; on the Plus plan it is unchanged.CTOWhen comparing subscription and API costs, account for the doubled limits.OpenAI to pool ChatGPT and Codex usage allowance from Decembersenior developerCodex and chat draw on one allowance, so schedule heavy agent tasks deliberately.tech leadRecalculate the team's subscription allowance, because Codex use will crowd out other ChatGPT use.Coding agents get spending limits as Cognition adds hierarchical budgets to Devintech leadSet budgets at team and task level, and define what happens to an interrupted agent task.Apple, Microsoft and Meta tighten internal AI usage limitsCTOEven large companies are keeping token costs in check, so plan how you will measure return on investment.senior developerExpect agent usage to be capped, and expect efficient prompting and context management to matter.
Machine analysis based on story titles and summaries, with a verification pass. The source of the facts is the article linked at each story.