PlanPulse
English

Recent changes

PlanPulse tracks new models, price changes, plan changes, and limit changes across AI APIs and products. Every entry carries a source and is human-reviewed before it lands here.

  1. New model OpenAI

    gpt-5.5 added with cached-input and long-context tiers

    Standard $5.00/1M input and $30.00/1M output, with separate short vs long-context pricing and Batch/Flex modes.
  2. New model Anthropic

    Claude Opus 4.8 and Sonnet 4.6 tracked with cache-read pricing

    Opus 4.8 at $5.00/$25.00 and Sonnet 4.6 at $3.00/$15.00 per 1M, both with 5m and 1h cache-write tiers.
  3. Price change DeepSeek

    deepseek-v4-flash listed at $0.14/1M input (cache miss)

    Cache-hit input drops to $0.0028/1M — a 50x discount — making it one of the cheapest 1M-context options.
  4. Plan change GitHub Copilot

    Copilot Pro plan separated into subscription + monthly AI credits

    $10/user/month with a monthly AI credit allowance; new sign-ups were temporarily paused on the checked page.
  5. Limit change Azure OpenAI

    GPT-5.5 Global flagged Needs Review pending region + deployment

    Headline token prices mirror OpenAI, but effective cost depends on deployment type and region quota.
  6. New product Cognition

    Devin Team plan added at $80/month minimum + $40/user

    Usage-based autonomous software-engineering agent with unlimited flex seats and a minimum monthly spend.
  7. New product App builders

    Lovable, Bolt, and Replit Agent added with usage models compared

    Credit-based (Lovable), token-based (Bolt $25/mo from 10M tokens), and effort-based (Replit Core $25/mo) builders.
  8. Limit change AWS Bedrock

    Bedrock kept as multi-model channel without a single normalized price

    Pricing is provider-, model-, and region-specific; a flat rate would be misleading, so Bedrock is treated as a channel rather than a single row.

How updates work

PlanPulse fetches pricing from each provider's official page, snapshots it, and diffs the new snapshot against the last one. Anything that moved becomes a candidate change.

High-risk fields — headline prices and usage limits — are then reviewed by a human before the change is published. That keeps the site reading as continuously updated without trading away accuracy: routine wording shifts pass through quickly, while the numbers that affect your bill get a second look first.