On June 12, 2026, Google locked Gemini 2.0 Flash out of its free API tier. The change appeared on Google’s official AI pricing page. It followed a wave of free tier limits that began in May. On June 15, Anthropic replaced flat-rate Claude access with a credit pool. GitHub Copilot confirmed usage-based billing on June 18. OpenAI added ads to its free ChatGPT tier and locked memory behind Plus. The common thread was simple. Free users were counting calories, not eating without limit. The changes hit developers who built on free quotas, students who relied on unlimited chat, and small teams that scaled without paying. The free lunch was over.
Not every user felt the cuts the same way. Free ChatGPT users saw ads appear between messages and lost persistent memory across sessions. Claude free users received a five-hour reset instead of four hours, but the ceiling dropped from 30 messages to 20. Gemini free users lost the 2.0 Flash model and saw daily request caps cut in half. Copilot users lost the $10 unlimited plan. The shift was not random. It tracked a broader industry move toward metered credit pools. For developers, the message was blunt. Build on free access and you build on sand. Vendor dashboards filled with quota warnings within days.
Why now? Compute costs remained high even as model prices fell. Providers spent 2025 subsidizing agentic coding, image generation, and long context windows. By June 2026, they wanted revenue. Open-source models from Qwen, Meta, and Mistral put pressure on API prices. Google answered with a 40% cut on paid Gemini Flash input prices. But free tiers shrank at the same time. That is the key contradiction. Paid customers won. Free users lost. The June changes made the trade explicit. Investors asked for margins, not user counts.
This report is not a how-to guide. It is a calorie count. We read the official announcements, checked pricing pages, and compared plan details. We found hard numbers. Anthropic’s free tier now resets every five hours with 20 messages per reset. GitHub Copilot free users get 2,000 completions per month. ChatGPT free users see ads unless they pay for Plus. Google killed a popular free API. Each change has a date and a source. We link to vendor pages and our prior coverage throughout. The all-you-can-eat AI era is over. Here is what replaced it.
How Do the Top Options Compare?
| Provider | June 2026 Change | Free Tier Limit | Paid Entry Price |
|---|---|---|---|
| Anthropic Claude | Ended agent subsidy, replaced flat access with credit pool | 20 messages per 5-hour reset | $20/month Pro |
| Google Gemini | Killed Gemini 2.0 Flash free API, cut Flash prices 40% | Daily cap reduced 50% on 3.5 Flash | $19.99/month Google AI Pro |
| GitHub Copilot | Usage-based billing replaces flat plan | 2,000 completions/month | $10/month base plus overages |
| OpenAI ChatGPT | Added ads and locked memory behind Plus | Basic chat with ads, limited Codex credits | $20/month Plus |
| Open-Source Models | No hosted API cap, but self-hosting required | Model weights free under license | Your GPU cost or cloud rental |
Prices and limits reflect official vendor pages as of June 20, 2026. Individual plan details may vary by region and account age.
1. Anthropic Claude , Best for tracking agentic credit consumption
Anthropic’s June 15, 2026 announcement confirmed the end of its agent subsidy. Flat-rate access for Claude coding disappeared. A credit pool replaced it. Users now receive a fixed monthly allocation, but agent tasks consume credits faster than standard chat. The official Anthropic announcement did not hide the logic. Subsidizing unlimited agent calls was not sustainable. Our report on the end of the subsidy detailed the new math. The credit system means a single long Claude Code session can drain a free allocation in hours.
The free tier changed too. Claude free users now reset every five hours instead of every four. That sounded generous, but the per-reset ceiling dropped from 30 messages to 20 for Claude Opus 4.8. The lower cap hit heavy users immediately. Paid Pro at $20 per month offered more credits but still no flat unlimited access. Team plans introduced overage fees above the credit pool.
The shift followed months of developer complaints about Claude Code limits. Many had built agent workflows on the free tier. After June 15, those workflows stalled. Some users moved to open-source alternatives. Others grudgingly paid. The change was not isolated. It aligned with a broader push by Anthropic to charge for agentic features. The all-you-can-eat era ended for Claude first.
Key strengths:
- ✅ Clear credit totals appear in the usage dashboard
- ✅ Five-hour reset reduces long lockouts
- ✅ Claude Opus 4.8 fast mode remains free for light use
- ✅ Official documentation explains credit burn rates
- ❌ Agent tasks consume credits much faster than chat
- ❌ Free ceiling dropped from 30 to 20 messages per reset
- ❌ No more flat-rate access for Claude Code
Who it’s for: Developers who need predictable credit tracking and can pay for heavy Claude Code sessions.
2. Google Gemini , Best for seeing price cuts on flash models
Google updated its official AI pricing page on June 12, 2026. The free API tier lost Gemini 2.0 Flash. Gemini 3.5 Flash remained free but with a daily request cap cut by 50%. The paid tier dropped 40% to $0.10 per million input tokens for Flash models. The Google AI pricing page showed the new numbers plainly. Our report on the Gemini 2.0 Flash shutdown captured the developer reaction. Free users suddenly saw quota errors on production calls.
Consumer free tier also shifted. Gemini Pro and Ultra moved fully behind Google AI Pro or Ultra plans. The free app now serves ads between responses. The free tier remains usable for light chat. It is no longer a development platform. Google’s paid price cut was the most aggressive move of June 2026. It put direct pressure on OpenAI and Anthropic.
But the free tier cuts offset the goodwill. Developers on X reported quota errors within hours. The free tier now covers only Gemini 3.5 Flash and a lighter Spark model. For free users, the story was simpler. Less access, more ads.
Key strengths:
- ✅ Paid Flash API prices dropped 40%
- ✅ Gemini 3.5 Flash still available free
- ✅ Pro plan includes more storage and priority
- ✅ Clear quota dashboard in Google AI Studio
- ❌ Gemini 2.0 Flash removed from free API
- ❌ Daily free request cap cut by 50%
- ❌ Ads injected into the free Gemini app
Who it’s for: Teams that need cheap high-volume API calls and can move to a paid Pro plan.
3. GitHub Copilot , Best for seeing usage-based coding costs
GitHub Copilot moved to usage-based billing on June 18, 2026. The old $10 per month flat plan no longer covered unlimited completions. Users received a monthly credit allowance, then paid for overages. The Agent multiplier made heavy coding sessions expensive. The official GitHub changelog documented the new math. Our report on Copilot usage-based billing broke down the credit costs. Many developers saw their bills jump within days.
Developers protested hidden costs. Some reported monthly charges rising from $10 to $60 after the switch. The Agent multiplier applied to multi-file edits and long-running tasks. GitHub added a dashboard, but the damage was done. The free tier still exists but now offers only 2,000 completions per month. Copilot Chat retained a small daily limit.
Many users moved to open-source alternatives like Qwen and Meta models. The message was clear. Free coding AI is no longer a reliable production tool.
Key strengths:
- ✅ Pay only for what you use
- ✅ Free tier still gives 2,000 completions per month
- ✅ Enterprise controls for credit alerts and caps
- ✅ GitHub API integration remains strong
- ❌ Overages can surprise you at month end
- ❌ Agent multiplier makes long tasks costly
- ❌ Flat $10 unlimited plan is gone
Who it’s for: Developers who track usage daily and can set hard spending caps to avoid overage bills.
4. OpenAI ChatGPT , Best for seeing free-tier ad and memory limits
OpenAI’s June 2026 changelog introduced ads in the free ChatGPT tier. Free users also lost persistent memory across sessions. Memory now required a Plus subscription. The image generation credit pool tightened for free users. The official OpenAI changelog listed the changes by date. Our coverage of the ChatGPT ad rollout detailed where ads appear. The free tier became a lead generation tool, not a full product.
ChatGPT Codex free tier appeared but with strict weekly agent credits. Users could run short agentic coding sessions, then hit a wall. The paid Codex plan started at $25 per month. A single repository review could consume the weekly credit. The old unlimited free ChatGPT experience was gone. Now every request counted.
OpenAI did not cut API prices as steeply as Google. GPT-5.2 remained free only for research preview. The company leaned into subscription revenue instead. Free AI image generators lost their top model. For users, the choice became simple. Tolerate ads and limits, or pay $20 per month for Plus.
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.
Key strengths:
- ✅ Free tier still offers basic chat
- ✅ Codex free credits for light use
- ✅ Plus plan has memory and no ads
- ✅ Clear usage dashboard for credits
- ❌ Ads interrupt free chat sessions
- ❌ Memory locked behind paywall
- ❌ Codex free credits and image generation are heavily capped
Who it’s for: Casual users who can tolerate ads and developers who need Codex with a paid plan.
5. Open-Source Models , Best for escaping usage caps entirely
The open-source route became the default escape hatch in June 2026. Qwen, Meta, Mistral, and other open-weight providers kept their model weights free. No API keys, no credit pools, no monthly caps. The trade was hardware. You run the model on your own GPU or rent compute. Our guide to the best free AI models with no API costs covered the top options. Developers who hit Anthropic or Google limits moved to these models within days.
Meta kept Llama 4 free for research and commercial use under its license. Qwen released coding models that matched closed tools on many benchmarks. Mistral maintained a free chat tier but pushed open-weight downloads. The pattern was clear. Open models gave users an exit from metered billing. But they required technical skill and infrastructure.
Not every user could make the switch. Consumers who only use chat apps cannot run a 70-billion parameter model. Small teams may not want to manage GPUs. Still, the pricing pressure from open weights was one reason Google cut paid API prices 40%. The all-you-can-eat era ended for hosted free tiers. It lived on for those willing to self-host.
Key strengths:
- ✅ Model weights free under open licenses
- ✅ No usage caps or credit pools
- ✅ No per-token billing when self-hosted
- ✅ Full control over data and latency
- ❌ You need your own GPU or cloud rental
- ❌ Setup and maintenance require technical skill
- ❌ Larger models may need expensive hardware
Who it’s for: Developers and teams who want to escape hosted API limits and can manage their own compute.
Frequently Asked Questions
Which AI providers changed their free tiers in June 2026?
Anthropic, Google, GitHub, and OpenAI all changed free access. Anthropic replaced flat Claude access with a credit pool on June 15. Google removed Gemini 2.0 Flash from its free API on June 12. GitHub Copilot moved to usage-based billing on June 18. OpenAI added ads and a memory paywall to ChatGPT free users.
Did Anthropic completely remove the Claude free tier?
No. The free tier still exists but with a five-hour reset and a lower ceiling of 20 messages per reset for Claude Opus 4.8. Agent tasks now consume credits. Flat-rate free access for Claude Code ended on June 15, 2026.
What happened to Google Gemini 2.0 Flash?
Google removed Gemini 2.0 Flash from the free API tier on June 12, 2026. Developers must now use Gemini 3.5 Flash free with a 50% lower daily cap or pay for Pro access. Paid Flash input prices dropped 40% to $0.10 per million tokens.
Is GitHub Copilot still free?
Yes, but the free tier now offers only 2,000 completions per month and a small daily Copilot Chat limit. The old $10 flat unlimited plan is gone. Usage-based billing started on June 18, 2026, with overage charges and an Agent multiplier.
Can free ChatGPT users opt out of ads?
No. OpenAI introduced ads in the free ChatGPT tier in June 2026. The only way to remove ads and regain persistent memory is to subscribe to ChatGPT Plus at $20 per month.
Where can I find the latest official pricing pages?
Check Google AI at ai.google, Anthropic at anthropic.com, GitHub at github.com, and OpenAI at openai.com. Each vendor posts changelogs and pricing updates. We link to these pages throughout our coverage.
Why did providers cut free tiers even as model prices dropped?
Compute costs remained high even as per-token prices fell. Providers spent 2025 subsidizing agentic coding, image generation, and long context windows. By June 2026, they prioritized revenue and margins. Open-source model pressure forced paid price cuts, but free tiers became a cost they were no longer willing to absorb.
What Should You Remember?
- Anthropic credit pools: Flat-rate Claude access ended on June 15, 2026. Agent tasks now consume credits.
- Google free cuts: Gemini 2.0 Flash API shut down on June 12, 2026. Free daily caps fell 50%.
- GitHub overages: Usage-based billing replaced the $10 unlimited Copilot plan. Watch the Agent multiplier.
- ChatGPT ads: Free tier now shows ads and locks memory behind Plus at $20 per month.
- Open-source shift: Developers looked to Qwen and Meta models as free limits tightened.
- Price war twist: Google cut paid API prices 40% even as free tiers shrank.
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.
