Changelog

Batch API now supports 100 texts per call, quality scores revamped.

• newBatch API limit increased from 25 → 100 texts per request.
• newQuality score now uses a 5-point scale with per-sentence granularity.
• improvedAggressive mode preserves 15% more factual nouns.
• improvedTokenizer updated for GPT-4o and Claude 3.5 token budgets.
• fixedCode fences inside triple backticks no longer get stripped.

LLMLingua mode (neural compression) available on Pro and Enterprise.

• newLLMLingua mode: neural compression that achieves 60–70% reduction with minimal quality loss.
• new/api/v1/stats/global endpoint — no auth required.
• improvedQuota check now returns X-RateLimit-Remaining and X-Quota-Remaining headers.
• fixedFixed edge case where markdown headers were stripped in balanced mode.

Model-aware tokenization, usage endpoint, Stripe billing.

• newmodel parameter — select gpt-4o, claude-3-5, or gemini for model-specific tokenization.
• newGET /api/v1/usage — per-key usage for the current billing period.
• newStripe subscriptions (Starter $19, Pro $99, Enterprise custom).
• improvedConservative mode now retains 98%+ semantic similarity (up from 95%).
• fixedCSRF token mismatch on Safari with strict SameSite settings.

Initial public release.

• newPOST /api/v1/compress — rule-based compression engine.
• newThree compression modes: conservative, balanced, aggressive.
• newFree tier: 50,000 tokens / month, no card required.
• newPlayground for interactive testing.
• newDashboard with API key management.