AI Limit Watcher

01 / RECENT CHANGES

• Jul 2026DeepSeek—↓[Pricing]DeepSeek V4 Flash input price $0.0938 → $0.0980 per 1M tokens (+4%)
• Jul 2026DeepSeek—↓[Pricing]DeepSeek V4 Flash output price $0.19 → $0.20 per 1M tokens (+4%)
• Jul 2026DeepSeek—↑[Pricing]DeepSeek V4 Flash input price $0.0980 → $0.0938 per 1M tokens (-4%)
• Jul 2026DeepSeek—↑[Pricing]DeepSeek V4 Flash output price $0.20 → $0.19 per 1M tokens (-4%)
• Jul 2026Claude—↑Adds access to a new model version for Pro users.
• Jul 2026GitHub Copilot—→[Pricing]Improved visibility and management of AI credit allocation for cost centers.

02 / MOST TIGHTENING — LAST 90 DAYS

6 tightening · 7 shifting · 17 steady

1. 1DeepSeek-67%
2. 2GitHub Copilot-44%
3. 3Tabnine-37%
4. 4ElevenLabs-32%
5. 5Mistral Le Chat-16%

03 / WHAT USERS REPORT

• “Blew past my five-hour limit after a single Claude Code prompt.” Claude source
• “Limits feel inconsistent and change without notice.” ChatGPT
• “Quota changed without prior notice and is very confusing.” Cursor
• “Free Gemini Pro access disappeared overnight.” Gemini
• “You have exceeded your premium request allowance.” GitHub Copilot
• “Bought an annual Pro subscription and the limits got quietly gutted.” Perplexity
• “The credit-to-quota switch changed how much I can actually do per day.” Windsurf
• “My credits now expire after six months — they didn't before.” Replit
• “The $300 Heavy tier is steep for what you get.” Grok
• “Cheap per-token, but the model lineup churns — V3.1, then V3.2-Exp, now V4, with deprecation dates each time.” DeepSeek
• “Video eats fast hours fast — about 8x an image job, so a Basic plan drains in no time.” Midjourney
• “Pinned eleven_multilingual_v1 in production — then it got a December removal date and I had to re-validate every voice.” ElevenLabs
• “I spent over $35 in a single day doing what previously cost me $20 for the entire month.” v0 source
• “ACUs cost $2.25 on the $20 plan, a hike from the $2 they cost on the $500-per-month subscription.” Devin source
• “Certain bots demand huge point quotas per message, making it impractical without a subscription.” Poe source
• “I don't know how many credits an operation will use before I run it, and credits are spent even on failed outputs.” Lovable source
• “Each failed fix attempt drains tokens without resolving the issue — easy to blow through a monthly allocation on one broken component.” Bolt.new source
• “Runway charges credits for failed or erroneous generations that need regenerating — you're charged for both attempts.” Runway source
• “Free songs can't be used commercially, and free users don't get the latest model.” Suno source
• “Credits shouldn't expire — at Kling they don't. If you keep your membership they should stack.” Luma Dream Machine source
• “Long-time users worry the Canva acquisition could shift focus from creators toward business customers.” Leonardo.ai source
• “Sora lead Bill Peebles: 'the economics are currently completely unsustainable.'” Sora source
• “The 'unlimited' messaging comes with a soft cap of ~150 messages/day under fair use.” Mistral Le Chat source
• “Topics that worked fine six months ago now trigger instant blocks, with unclear boundaries.” Character.AI source
• “200 Creator credits vanished in a single Avatar IV session — every render iteration counts, not just the final video.” HeyGen source
• “Re-renders consume additional minutes from the monthly allowance; production costs exceed advertised pricing.” Synthesia source
• “Teams that previously ran Plus + AI for $18/user now must jump to Business at $20/user for AI.” Notion AI source
• “Tabnine discontinued their Dev plan and pivoted to enterprise-only pricing.” Tabnine source
• “Free tier provides only 10 slow credits/week (~40 images), with no priority-queue access.” Ideogram source

We track whether these changes are real.

04 / EXPLORE

See exactly what changed in each AI tool →