On June 15, 2026, Anthropic confirmed a major billing shift. The company ended its agent subsidy for Claude Pro and Max subscribers. Flat-rate access to Claude Code and agentic workflows stopped. A credit pool replaced it. Free users faced a lower daily message cap. The official change appeared on Anthropic’s pricing page. Anthropic said the old model paid for agent compute out of subscription fees. That subsidy became too expensive. The new credit pool forced users to track every tool call. Anthropic ends agent subsidy June 15 credit pool replaces flat-rate access covered the initial details. Users who relied on long Claude Code sessions lost the most. The free tier now resets on a five-hour timer. It no longer supports heavy agent use. Paid users must now buy credits or watch their balance fall with every file edit. The change arrived with little warning. Users discovered the new limits when their sessions stopped mid-task. Anthropic said the decision was necessary to keep Claude sustainable. Free users heard a different message. The best free agent access was gone. Many free users had built daily workflows around Claude Code. Those workflows broke on June 15.
Google made its own cuts on June 2, 2026. The free Gemini API lost access to Gemini 2.0 Flash. Pro model endpoints moved behind paid billing. Free users could only call Gemini 3.5 Flash. Google published the change on its AI site. The move followed months of rising inference costs. Agentic calls are expensive because they chain many model requests. A single task can burn through fifty or more calls. Gemini free tier cuts 2026 reported the new quotas. Free users saw request limits drop. Some developers reported quota errors within hours. The free tier became a sandbox, not a workbench. Google’s free tier still exists, but it is smaller. The days of free Pro model access are over. Third-party tools that wrapped Gemini 2.0 Flash also broke. Developers scrambled to swap models or add billing. Google’s move was not as abrupt as Anthropic’s, but it hurt the same group. Free users who wanted agentic workflows had nowhere to go. The free API tier now serves only light experimentation. Heavy use requires a credit card.
OpenAI and GitHub also shifted billing. On June 9, 2026, OpenAI reduced free Codex access. The free tier now runs ads in some regions. ChatGPT free users get 25 messages per five hours for agentic coding. GitHub Copilot moved to usage-based billing in June 2026. Developers faced a multiplier on premium model requests. GitHub Copilot usage-based billing June 2026 documented the change. AI API free tiers limits 2026 showed the broader pattern. The billing crisis was not one vendor. It was a structural reset. OpenAI’s ad test and Copilot’s multiplier are two sides of the same coin. Free users now face a paywall for autonomous work. Chat remains available, but agent features are limited. OpenAI said ads help keep free chat running. The company did not explain how ads help agent users. Copilot’s multiplier drew immediate complaints on developer forums. Both changes share a common cause. Agent inference is expensive. Someone has to pay.
Why does this matter? Agentic AI costs more than chat. Each agent step calls a model, a tool, and a planner. Stanford HAI’s AI Index has tracked rising inference costs. Free users once absorbed these costs through venture subsidies. Now vendors are passing them back. The result is a smaller free tier. Users must watch quotas, timeouts, and credit drains. Agentic AI billing crisis free users 2026 explained the economics. Free users are not just losing features. They are losing the ability to run autonomous tasks at all. The free tier is becoming a demo. The paid tier is becoming the only real tool. Developers who built free agent workflows on these APIs must now migrate or pay. The shift also changes the open-source conversation. More users are looking at local models. But local models have their own hardware costs. Free does not mean free anymore. This is the new math of agentic AI.
How Do the Top Options Compare?
| Platform | Free Tier Change | New Limit | Effective Date |
|---|---|---|---|
| Anthropic Claude | Flat-rate agent access ended; credit pool introduced | Five-hour reset; lower daily message cap | June 15, 2026 |
| Google Gemini | Gemini 2.0 Flash API shut down; Pro models paid | Free calls limited to Gemini 3.5 Flash | June 2, 2026 |
| OpenAI ChatGPT | Free Codex access reduced; ads in free tier | 25 messages per 5 hours for Codex | June 9, 2026 |
| GitHub Copilot | Usage-based billing; free tier reduced | 2,000 completions per month | June 2026 |
| Mistral Le Chat | No credit pool; free tier mostly unchanged | Free chat with limited tool use | June 2026 |
Free AI News may earn a commission if you sign up for a paid plan through links on this site. This does not affect our reporting. Limits and dates are based on vendor announcements as of June 2026. Check provider pages for current terms.
1. Anthropic Claude , Long agent sessions on a budget
Anthropic’s June 15 change hit free and paid users. The flat-rate era ended. Anthropic moved agent usage to a credit pool. Free tier messages now reset on a five-hour timer. Paid subscribers lost unlimited Claude Code access. The old plan covered agent compute from $20 and $200 monthly fees. The new plan charges credits per tool call. A single long coding session can drain a day’s credits. Anthropic agent billing split June 2026 detailed the new system. Users who want predictable agent access now need the Max plan. Even then, the credit cap applies. For free users, the cap bites fast. They must choose between chat and agent runs. Many will stick with plain Claude messages.
Others will move to open-source models. The change also affects API users who built on Claude Code. They now need a billing account and a credit balance. Anthropic said the old subsidy was unsustainable. The company is not wrong about costs. But the change removes a popular free path to agentic coding. Free users who loved Claude Code now face a hard stop. The five-hour reset means a user can try again later, but cannot finish a long task. That is the new reality for free Claude users.
Key strengths:
- ✅ Credit tracking is transparent
- ✅ Free chat access remains available
- ✅ Paid users can still access Claude Code
- ✅ Five-hour reset is predictable
- ❌ No unlimited agent use on any plan
- ❌ Credits drain quickly on multi-step tasks
- ❌ Free tier cap is too low for agent work
Who it’s for: Anyone who wants Claude agent access but must watch every tool call.
2. Google Gemini Free Tier , Light experimentation and single-step prompts
Google’s free Gemini tier narrowed sharply on June 2, 2026. Google AI confirmed that Gemini 2.0 Flash API endpoints would stop serving free requests. Pro model endpoints now require a paid key. Free users are limited to Gemini 3.5 Flash. That model is faster but less capable for complex agent loops. The change followed a broader quota crackdown. Gemini free tier cuts 2026 reported the new limits. Developers who built free tools on Gemini 2.0 Flash had to migrate or pay. Google framed the move as a quality control step. Free users saw it as a paywall.
The free tier remains useful for chat. It is no longer useful for autonomous agents. Multiple developers reported seeing HTTP 429 errors after just a few agent-style requests. The free quota reset every minute, but the total call volume was too low. Google’s paid plans now offer more generous limits, but the free tier lost its former power. For users who only need a quick answer, Gemini 3.5 Flash still works. For anyone building an agent, the free tier is effectively closed. Google has not restored Gemini 2.0 Flash access. It likely will not.
Key strengths:
- ✅ Gemini 3.5 Flash is still free
- ✅ Paid Pro access is available
- ✅ Clear API documentation
- ✅ Fast response times
- ❌ Gemini 2.0 Flash API is gone
- ❌ Pro models require a paid plan
- ❌ Free quota errors appear quickly
Who it’s for: Casual users who need a fast free chatbot but not agent workflows.
3. OpenAI ChatGPT Free Tier , Casual users testing ChatGPT and light Codex
OpenAI tightened the free tier on June 9, 2026. Free Codex access shrank to 25 messages per five hours. Ads began appearing in some free tier sessions. OpenAI said the changes support continued free access. The company pointed to high agent inference costs. The Codex limit is the real story. A complex coding agent can burn 25 messages in minutes. That makes free Codex a trial, not a tool. OpenAI still offers a paid ChatGPT plan. The free tier now pushes users toward it. For developers, the message is clear. Free agentic coding is over.
The ads are a separate test. OpenAI has not said which regions will see ads first. But the combination of ads and limits changes the free tier experience. Users who relied on free Codex for small scripts can still do a few runs. They cannot complete a multi-file project. The paid ChatGPT plan includes more Codex messages, but even that has a cap. OpenAI has made the free tier a preview. The full product now costs money. Free chat remains, but free agents are gone.
Key strengths:
- ✅ ChatGPT chat remains free
- ✅ 25 Codex messages per five hours is something
- ✅ Paid upgrade is straightforward
- ✅ Transparent message cap
- ❌ Ads in free tier reduce usability
- ❌ 25 messages is too few for real coding
- ❌ Free Codex resets on a five-hour timer
Who it’s for: Users who want occasional ChatGPT access and can tolerate ads and limits.
4. GitHub Copilot Free Tier , Developers on a strict budget
GitHub Copilot moved to usage-based billing in June 2026. The free tier dropped to 2,000 completions per month. Premium model requests now carry a multiplier. GitHub published the new pricing on its changelog. Developers who used Copilot for long agent sessions faced hidden costs. The change sparked immediate backlash. GitHub Copilot usage-based billing June 2026 reported the developer outcry. A single multi-file refactor can exhaust the monthly free allowance in one afternoon. Free users must choose between completions and chat. Many are switching to local models.
The free tier now works only for small projects. GitHub said the old flat-rate plan could not sustain agentic coding costs. The company introduced a usage dashboard to track remaining completions. But the multiplier on premium models makes cost prediction hard. Developers reported seeing charges for requests they did not realize were premium. The free tier now feels like a trial. Paid plans remove the cap, but the multiplier remains. For open-source maintainers, this is a painful change.
Key strengths:
- ✅ 2,000 free completions per month
- ✅ Clear usage dashboard
- ✅ Paid plans remove the cap
- ✅ GitHub integration remains strong
- ❌ Multiplier on premium models adds surprise costs
- ❌ Free allowance drains quickly with agent features
- ❌ Developer backlash over opaque billing
Who it’s for: Developers who need light code completion and can track monthly usage.
5. Mistral Le Chat Free Tier , Users seeking a free assistant without agent limits
Mistral Le Chat kept its free tier mostly intact through June 2026. Mistral AI did not introduce a credit pool for agent use. Free users can still access Le Chat and Mistral Small. The company has open-weight models available on Hugging Face. That matters for free users burned by Anthropic, Google, and OpenAI. Mistral’s free tier is not unlimited, but it has no usage multiplier. Users who want to run agents locally can download Mistral models. This path has hardware costs, but no per-tool-call billing. Mistral has not been immune to cost pressure. The company raised some API prices for commercial users. But its free chat tier remains a real option. For many free users, that is enough.
Key strengths:
- ✅ No credit pool for free chat
- ✅ Open-weight models available
- ✅ No usage multiplier on free tier
- ✅ Works for local agent experiments
- ❌ Free tier is not unlimited
- ❌ Local models require hardware
- ❌ Fewer managed agent features
Who it’s for: Free users who want a stable chat tier or local model path after major providers cut access.
Frequently Asked Questions
What happened to Anthropic's free agent access on June 15, 2026?
Anthropic ended flat-rate agent access for paid subscribers and lowered the free tier message cap. Agent usage now runs through a credit pool, so free users cannot run long Claude Code sessions without hitting limits. Paid users also lost unlimited Claude Code access.
Did Google remove free Gemini API access?
Yes. On June 2, 2026, Google shut down free access to Gemini 2.0 Flash. Free API users are limited to Gemini 3.5 Flash, while Pro model endpoints require a paid key. Third-party tools that used Gemini 2.0 Flash broke immediately.
How many free Codex messages do ChatGPT users get now?
OpenAI set the free Codex limit at 25 messages per five hours starting June 9, 2026. Ads also appear in some free tier sessions. A complex coding task can burn through that allowance in minutes.
What is GitHub Copilot's new free tier limit?
GitHub Copilot now offers 2,000 completions per month on the free tier. Premium model requests carry a usage multiplier, which can add hidden costs. A single multi-file refactor can exhaust the monthly allowance.
Why are AI vendors cutting free agent access?
Agentic AI calls are expensive because each task chains multiple model and tool requests. Vendors are replacing flat-rate subsidies with usage-based billing to control inference costs. The old free tiers were unsustainable at scale.
Can free users still run AI agents at all?
Yes, but with tight limits. Free tiers now function as trials or sandboxes. Heavy autonomous work generally requires a paid plan or a local open-source model. Some providers, like Mistral, still offer free chat without credit pools.
What Should You Remember?
- Anthropic billing shift: Flat-rate Claude agent access ended June 15, 2026.
- Google free API cut: Gemini 2.0 Flash shut down for free users June 2.
- OpenAI Codex cap: Free users get 25 Codex messages per five hours.
- GitHub Copilot change: Free tier now 2,000 completions per month.
- Agent costs drive change: Multi-step tool calls make agent workflows expensive.
- Free tier is a demo: Heavy autonomous work now requires paid plans.
Free AI News is an independent editorial publication. Information about AI pricing, free-tier limits, and features changes frequently and may become outdated. Always verify current details through the vendor’s official pages. Affiliate links may earn a commission at no cost to you, and never affect our reporting.
