Free AI Pricing Changes June 2026: Google, Anthropic Cut Free Tiers

On June 12, 2026, Google confirmed that the free API tier for Gemini 2.0 Flash would end immediately. Developers who used the model for testing, prototypes, and lightweight applications lost access unless they moved to Gemini 3.5 Flash or switched to a paid compute plan. The change hit hobbyists and small automation projects first. Free AI News tracked the update through Google Gemini API free tier tightened and Google AI’s official pricing page. The cutoff arrived just weeks after Google advertised lower subscription prices. That sequence made the free-tier reduction easy to miss. Many users saw a discount on paid plans and assumed free access was safe. It was not.

Anthropic followed with a June 15, 2026 policy change that removed flat-rate access to Claude agents. The company replaced the old system with a monthly credit pool that applied to Claude Code, OpenClaw, and other agent workflows. Free users received a smaller pool that reset every five hours, while paid users got a 50 percent higher limit in Claude Code. The move ended what Anthropic called an agent subsidy. The details appeared on Anthropic’s official pricing page and in our earlier report on the Anthropic agent billing split. The shift meant that long-running agent tasks could burn through credits quickly. Free users could no longer rely on continuous Claude Opus 4.8 access.

OpenAI and Microsoft adjusted their free tiers in the same window. On June 9, 2026, ChatGPT free users in the United States and United Kingdom began seeing ads after their first three responses. OpenAI also launched a Codex free tier with agentic coding, but capped daily runs at ten and then cut the limit to five after a demand spike. Microsoft put some Office apps behind a Copilot paywall and reworked GitHub Copilot into usage-based billing on June 2, 2026. Developers reported surprise charges when premium model requests multiplied token counts. The combined changes hit individual developers, students, and small teams hardest.

Why did the free tiers shrink in June 2026? The business context points to agentic AI costs. Agent workflows consume many more tokens than a single chat reply. Providers gave away free agent access in early 2026 to attract developers, then discovered that heavy usage became a cost problem. Google cut subscription prices in May, but by June the company was tightening free compute quotas. Anthropic moved to credits. GitHub Copilot added usage multipliers. The result was a split: cheaper paid headlines and tougher free limits. Users who relied on free agent tools faced hard choices.

How Do the Top Options Compare?

Limits were accurate as of June 25, 2026, based on vendor pricing pages. 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.

1. Google Gemini Free API , Developers testing Gemini models on a small scale

The most visible June 2026 change from Google was the shutdown of the free API for Gemini 2.0 Flash on June 12. Before the change, free developers could send up to 15 requests per minute to the model. After the change, the rate limit dropped to zero. Google pointed users to Gemini 3.5 Flash, which remained available through a free compute quota. That quota reset daily and allowed only 2 requests per minute. The move forced migration for thousands of small projects. Google paired the free API cut with subscription price reductions. On June 4, Google AI Ultra dropped from $59.99 to $39.99 per month. The company framed the shift as a tradeoff: lower prices on paid plans, but fewer free allowances. The official pricing page at Google AI listed the new compute quota. Many developers felt the tradeoff was worse for experimentation. The Gemini 2.0 Flash shutdown report captured the backlash. Free tier users who wanted more compute had to enter a credit card and buy a paid tier. That conversion push was deliberate. Google saw agentic coding and image generation consume compute at high rates. The company also tightened free access to Pro models through the API. Existing free users received a 14-day grace period, but after June 26, 2026 the old limits were gone for good.

Key strengths:

• ✅ Gemini 3.5 Flash remains free through a daily compute quota
• ✅ Google AI Ultra price dropped by $20 per month
• ✅ Clear migration path from 2.0 Flash to 3.5 Flash
• ❌ Free API for Gemini 2.0 Flash ended with zero requests per minute
• ❌ Free compute quota is easy to exhaust at 2 requests per minute
• ❌ Grace period lasted only 14 days

Who it’s for: Developers who want limited free access to Gemini 3.5 Flash and can pay when compute needs grow.

2. Anthropic Claude Credit Pool , Teams that want predictable monthly agent usage

On June 15, 2026, Anthropic replaced flat-rate access to Claude agents with a credit pool. The change applied to Claude Code, OpenClaw, and other agent features. Free users no longer received endless agent runs. Instead, they got a smaller credit pool that reset every five hours. Paid users received a 50 percent larger Claude Code limit than free users. The new system ended what Anthropic had called an agent subsidy. The official announcement and pricing details were visible on Anthropic’s site. The credit overhaul also changed how Claude Opus 4.8 was billed. Anthropic kept the model at the same base price, but introduced a cheaper fast mode and a free tier for short answers. The free tier was limited to three messages per five-hour window. Our report on the Claude credit overhaul broke down the new math. For heavy agent users, the monthly bill rose. For light users, the credit pool reduced waste. Free users felt the squeeze most. A typical Claude Code session could consume the free pool in under an hour. The five-hour reset helped, but long tasks required paid credits. Anthropic also changed console credit policies. New accounts received $1 in free console credits instead of the previous $5. That reduction made it harder to test agent workflows before paying.

Key strengths:

• ✅ Paid users get a 50 percent higher Claude Code limit
• ✅ Credit pool prevents flat-rate waste for light users
• ✅ Opus 4.8 fast mode is cheaper than standard mode
• ❌ Free tier loses flat-rate agent access
• ❌ Free console credits dropped from $5 to $1
• ❌ Long agent sessions burn through the free pool quickly

Who it’s for: Paid Claude subscribers who want predictable agent limits and can afford monthly credits.

3. ChatGPT Free Tier and Codex , Casual users who can tolerate ads and daily limits

OpenAI introduced ads to the ChatGPT free tier on June 9, 2026. Free users in the United States and United Kingdom began seeing advertisements after their first three responses in a session. The ads appeared as small cards below answers. OpenAI said the change would support free access. Users could remove ads by upgrading to ChatGPT Plus for $20 per month. The rollout matched earlier reporting on ChatGPT free tier ads. At the same time, OpenAI launched a Codex free tier with agentic coding. The free tier allowed ten agentic runs per day at launch. Demand spiked within the first week. OpenAI cut the limit to five runs per day on June 16, 2026. Free Codex also restricted repository size to 25 files and blocked private repository use. The change gave developers a taste of Codex, but not a usable daily driver. ChatGPT memory on the free tier also changed. OpenAI limited free memory to two stored facts per session, down from ten. The memory feature still worked, but free users lost long-term context faster. The official details appeared on OpenAI’s site. For casual users, the free tier remained usable. For power users, the limits pushed them toward paid plans.

Key strengths:

• ✅ Codex free tier introduced agentic coding at no cost
• ✅ Free users can still access ChatGPT with ads
• ✅ Memory feature remains free with smaller context
• ❌ Ads after three responses in US and UK markets
• ❌ Codex free runs cut from ten to five per day
• ❌ Free memory reduced from ten facts to two per session

Who it’s for: Casual users who need basic ChatGPT access and occasional Codex runs.

4. GitHub Copilot Usage-Based Billing , Developers who write code occasionally and monitor usage

GitHub moved Copilot to usage-based billing on June 2, 2026. The old flat-rate plan disappeared for new users. Free users retained a base allowance of 2,000 completions per month. Premium model requests carried a multiplier. A single premium request could count as five standard completions. Developers who used Copilot’s agent mode reported surprise charges. The GitHub Copilot usage-based billing report documented the shifts. The change hit heavy users immediately. A developer who ran Copilot’s agent mode for a day could burn through a week of standard quota. GitHub argued that usage-based billing was fairer for light users. The company pointed to its official pricing page on GitHub. Many developers disagreed, especially those who had signed up for a flat $10 monthly plan. Free users also lost some premium completions. GitHub limited free Copilot to the base model. Copilot Pro, priced at $10 per month, included premium model access but still used multiplier billing. The hidden costs sparked a backlash. One developer reported a $38 monthly bill after expecting a $10 flat fee. The free tier remained valuable for occasional use, but unpredictable costs undermined trust.

Key strengths:

• ✅ Free tier keeps 2,000 completions per month
• ✅ Light users can avoid flat subscription fees
• ✅ Base model completions stay free
• ❌ Premium requests carry a multiplier up to five times
• ❌ Flat $10 plan no longer covers heavy usage
• ❌ Usage tracking is hard to predict

Who it’s for: Developers who use Copilot occasionally and can monitor token consumption.

Frequently Asked Questions

What Should You Remember?

• Google Gemini free API: Gemini 2.0 Flash free requests went to zero on June 12, 2026.
• Anthropic credit pool: Flat-rate Claude agent access ended June 15, replaced by credits.
• ChatGPT ads: Free users in the US and UK saw ads after three responses starting June 9.
• GitHub Copilot billing: Usage-based billing began June 2, with premium request multipliers.
• Codex free tier: Agentic coding launched free, then daily runs dropped from ten to five.
• Paid price cuts: Google AI Ultra dropped $20 per month, but free limits tightened.
• Affiliate note: Free AI News may earn a commission if you sign up for paid plans through links on this site.

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.