AI pricing week: DeepSeek, Google, GitHub all moved at once

Three pricing changes landed in the last week of May 2026 that, taken together, shift what most businesses will spend on AI this year.

DeepSeek made its 75% V4 Pro discount permanent. Google launched a $100/month AI Ultra tier and cut the old $250 plan to $200. GitHub Copilot switches to usage-based billing on June 1, which means the flat-rate ride on premium requests ends.

Individually, each is a pricing story. Together, they point to where the market is going: tokens are getting cheap for basic work, expensive for frontier agent tasks, and confusing in the middle.

DeepSeek V4 Pro: $0.87 per million output tokens, permanently

DeepSeek launched V4 Pro in April with a promotional 75% discount. The promotion was supposed to expire May 31. Instead, DeepSeek made it permanent on May 23, according to reporting from The Next Web and Engadget.

The new pricing: $0.003625 per million input tokens and $0.87 per million output tokens. Down from $0.0145 input / $3.48 output.

For comparison, output token prices per million:

DeepSeek V4 Pro now costs less than Google's cost-optimized Flash model. It supports a one-million-token context window at that price, which matters for document analysis, legal review, and code comprehension where input tokens pile up fast.

The catch: DeepSeek is a Chinese AI company, and routing sensitive workloads through it carries geopolitical and compliance risk that varies by industry. Anthropic has publicly accused DeepSeek of "distillation attacks," claiming DeepSeek trained on Claude's outputs to improve its own models. DeepSeek has not addressed this in detail. Salesforce, for reference, projects $300 million in Anthropic token spending this year. At DeepSeek's prices, an equivalent volume would cost a fraction of that. Whether the savings justify the risk is a question each company answers for itself.

Google AI Ultra: new $100 tier, Gemini Spark agent

At Google I/O on May 19, Google restructured its AI subscription lineup. The headline: a new $100/month AI Ultra tier, as Google's announcement lays out.

What the $100 tier includes:

• 5x the usage limits of the $20 AI Pro plan, in both Gemini and Google Antigravity
• 20TB of cloud storage
• YouTube Premium individual subscription
• Priority access to Google Antigravity (the agent development platform)
• Access to Gemini Spark, a 24/7 AI agent that runs on Google Cloud VMs and connects to Gmail, Calendar, Docs, and Slides

The existing top-tier plan dropped from $250 to $200/month. It keeps 20x Pro usage and adds Project Genie, an experimental world-building tool. Both Ultra tiers include Gemini Spark, which started rolling out to U.S. beta subscribers the week of May 25.

According to Android Police, the pricing targets ChatGPT Pro ($200/month) and Claude Max ($200/month). Google also moved from daily prompt caps to a compute-based usage model. Your quota now depends on how complex each prompt is, not just how many you send. Limits refresh every five hours up to a weekly cap.

For businesses already paying for Google Workspace, the $100 Ultra tier bundles enough storage and YouTube Premium that the actual AI cost feels closer to $60-70/month.

GitHub Copilot: usage-based billing arrives June 1

GitHub Copilot flips to AI Credits on June 1, as UsageBox details. The old "premium requests" bucket disappears. In its place: a dollar-for-dollar credit system where 1 credit equals $0.01.

The tiers:

• Copilot Pro ($10/month): $10 in credits included
• Copilot Pro+ ($39/month): $39 in credits included
• Copilot Business ($19/seat/month): $19 in credits per seat

Code completions and Next Edit suggestions remain free. The credits only get consumed by Chat, Agent Mode, multi-file edits, and code review, the features that run on expensive frontier models.

The math gets sharp fast. A Copilot Pro+ user who runs five Agent Mode tasks per day on Claude Opus 4.7, each around 30K tokens, burns roughly $5-8 in credits. Two large refactors on GPT-5.5 across multiple files can add another $4-6. That is $10-14 per day. The $39 monthly allowance covers three or four days of that pattern. After that, you pay overage.

The old premium-request system averaged costs across all users. Heavy Agent Mode users were effectively subsidized by everyone else. The new system makes the per-task cost visible, which is more honest but will surprise teams that have grown reliant on agent workflows.

If your team mostly uses tab-completion with occasional Chat, the change is invisible. If your developers are running Agent Mode all day, budget for it now.

What this means for your AI spend

Three pricing shifts in one week, and the direction depends on what you use AI for.

For API consumers doing document processing, form parsing, or intake automation: DeepSeek's permanent pricing makes high-volume, long-context work much cheaper. The question is whether your compliance team signs off.

For teams building with Google's stack: the $100 Ultra tier is the first agent-capable plan priced below $200. Gemini Spark is still in beta and U.S.-only, but the price point opens agent access to smaller teams that could not justify $250/month.

For development teams on Copilot: pull your premium-request usage from the billing dashboard before June 1. Figure out which developers are burning the most credits. Decide now whether to upgrade, switch models, or move to a competitor like Cursor that still offers fixed-cost plans.

The pattern underneath: basic AI capability is getting cheaper fast. Agent-grade capability is getting more expensive because it burns more compute. The gap between the two is widening, and pricing models are starting to reflect that split.

---

Sources:

• The Next Web: DeepSeek made its 75% discount permanent (May 2026)
• Google Blog: Everything new in our Google AI subscriptions (May 19, 2026)
• Android Police: Google's new $100 Gemini plan (May 19, 2026)
• Digital Trends: Google AI subscriptions new $100 tier (May 19, 2026)
• UsageBox: GitHub Copilot usage-based billing (May 2026)