What's new in Anvat.

Tier-7: OpenAPI/Postman API reference, weekly cost-efficiency leaderboard, segment landing pages

Three big developer-facing surfaces shipped: a complete /api-reference page backed by a downloadable OpenAPI 3.1 spec + Postman v2.1 collection (importable into any API tool), a weekly auto-updated /leaderboard ranking frontier LLMs by intelligence-per-dollar across 6 benchmark categories with Dataset JSON-LD for AI answer engines, and three segment landing pages /for/startups, /for/agencies, /for/researchers with Service + FAQ JSON-LD targeting high-intent segment search.

• •New: /api-reference — full endpoint docs, downloadable openapi.yaml + Postman v2.1 collection
• •New: /openapi.yaml + /anvat-postman-collection.json static downloads
• •New: /leaderboard — weekly cost-per-intelligence ranking across coding/agentic/reasoning/math/long-context/multimodal with Dataset JSON-LD
• •New: /for/startups, /for/agencies, /for/researchers — segment-specific selling with Service + FAQ schema
• •Updated: nav (added Top + API tabs), footer (For-teams column + leaderboard link), tools hub (leaderboard + API reference)
• •Updated: middleware (allow .yaml/.json bypass for spec downloads), sitemap, llms.txt all wired

Tier-6: interactive Opus 4.8 demo, video walkthroughs, per-country pricing, Mythos tracker

Four new surfaces shipped: a bring-your-own-key in-browser Opus 4.8 audit demo that runs against api.anvat.app directly, a video walkthroughs hub with VideoObject JSON-LD and full transcripts, five localized per-country pricing pages (Vietnam, India, Brazil, Philippines, Indonesia) with currency conversion + local payment methods + tax notes + FAQ schema, and a Mythos vs Opus 4.8 benchmark tracker that auto-updates when Anthropic publishes more numbers.

• •New: /demo/audit — bring-your-own-key Opus 4.8 audit/refactor/explain/test demo, streaming SSE in-browser
• •New: /videos hub + /videos/zcash-opus-4-8-walkthrough with VideoObject JSON-LD + full transcript
• •New: /pricing/{vietnam,india,brazil,philippines,indonesia} — localized pricing with VND/INR/BRL/PHP/IDR display + local payment + tax + FAQ JSON-LD
• •New: /blog/mythos-benchmark-tracker — living comparison of Mythos vs Opus 4.8
• •Updated: /pricing main page links to all 5 country pages
• •Updated: llms.txt + sitemap + middleware whitelist all new routes

Tier-5 content blitz: 6 blog posts, 2 new model pages, /benchmarks tool

Total-force coverage of the trending AI cycle — GPT-5.6 release tracker, Gemini 3.5 Pro tracker, Claude Skills open standard guide, DeepSeek V4 Pro deep-dive, EU AI Act enforcement guide, Cursor vs Antigravity vs Devin IDE comparison. Added Gemini 3.5 Flash + DeepSeek V4 Flash model pages, updated DeepSeek V4 Pro with correct 1M context + actual benchmark numbers, and shipped a new interactive /benchmarks comparison tool spanning Claude, GPT, Gemini, and DeepSeek across coding, agentic, reasoning, math, and long-context dimensions.

• •New: /blog/gpt-5-6-release-tracker — iris-alpha codename, Codex log leak, Polymarket odds
• •New: /blog/gemini-3-5-pro-tracker — June launch + Flash benchmark deep-dive
• •New: /blog/claude-skills-open-standard — 32+ AI tools support SKILL.md format
• •New: /blog/deepseek-v4-pro-deep-dive — 80.6% SWE-Bench at $3.48/MTok output
• •New: /blog/eu-ai-act-2026-enforcement-guide — August 2 deadline + Code of Practice
• •New: /blog/cursor-vs-antigravity-vs-devin-2026 — three-way IDE shootout
• •New: /models/gemini-3-5-flash — Google's new fast tier
• •New: /models/deepseek-v4-flash — 13B-active open-weight tier
• •New: /benchmarks — interactive comparison of 8 models across 12 benchmarks
• •Updated: /models/deepseek-v4 with full benchmark table + 1M context correction
• •Updated: 9 new news digest entries (GPT-5.6 leak, Antigravity, Cursor 3, EU AI Act, etc.)

Coverage: Claude Mythos + Zcash bug + AI-assisted security research

Three timely deep-dives published — Claude Mythos / Project Glasswing explainer, how Opus 4.8 found the four-year-old Zcash zero-knowledge proof bug, and a practical guide to AI-assisted security research workflows. New /use-cases/security-research page positions Anvat for the security-research workflow that produced the Zcash discovery.

• •New: /blog/claude-mythos-explained — Project Glasswing, benchmark gap vs Opus 4.6, $25/$125 partner pricing, GA timeline
• •New: /blog/zcash-bug-claude-opus-4-8 — full timeline of the May 29 discovery, workflow that found it, what it proves
• •New: /blog/claude-security-research-2026 — patterns that ship findings, model selection per audit task, what Glasswing changes
• •New: /use-cases/security-research — quickstart code, FAQ, related models/blog

Blog, comparison pages, model catalog, and cost calculator

Major content drop expanding the marketing surface — five evergreen blog posts, four LLM gateway comparisons, four per-model spec pages, four use-case setup guides, and an interactive cost calculator with shareable URLs.

• •New: /blog with 5 launch posts (Claude pricing, Claude Code vs Cursor, custom base URL setup, OpenRouter alternatives, cheap Claude API)
• •New: /vs/{openrouter,litellm,portkey,helicone} with structured decision matrix per competitor
• •New: /models/{claude-opus-4-8,claude-sonnet-4-6,claude-haiku-4-5,gpt-5} with pricing tables, benchmarks, code samples
• •New: /use-cases/{claude-code,cursor,coding-agents,rag} with quickstart guides
• •New: /cost-calculator with sliders, presets, shareable URL hashes
• •Improved: full hreflang + JSON-LD + dynamic OG image coverage on every new page

Multi-region SEO upgrade for Western and Chinese markets

Reworked metadata, hreflang alternates, and verification flow to maximise discoverability across Google, Bing, Yandex, Baidu, and Sogou. PWA manifest, Apple mobile-web-app meta, and Open Graph alternate locales now ship on every page.

Telegram contact widget + mobile bottom nav redesign

Added a floating Telegram contact button (`t.me/anvat_app`) site-wide. Redesigned the mobile bottom navigation with curated tabs + a More bottom sheet for secondary items, fixing the FAB-vs-nav overlap reported by users on small screens.

Stripe live mode — production billing online

Promoted Stripe configuration to live keys across the gateway and frontend. Card and crypto checkout (USDC, USDT) now operational for prepaid credit packs and subscriptions.

Anti-spam quota system with rolling spend caps

Tightened per-tier RPM limits, added concurrent request enforcement, and introduced rolling 5-hour + 7-day spend caps to prevent abuse without affecting legitimate workloads. Admin accounts bypass all quotas.

Tier-3 referral commission — paid on every package purchase

Subscription purchases now pay 10% lifetime commission to the referring user (Tier 3 of the referral programme). Tier 1 (signup bonus) and Tier 2 (referrer reward) unchanged. Commission ledger added to /app/referrals.

Referral cap of 10 per account

Capped the number of referrals an individual account can claim at 10 to keep the programme aligned with our cost model. Existing accounts with more than 10 referrals are grandfathered.

Claude Opus 4.8 + Sonnet 4.6 1M-context tier

Added Claude Opus 4.8 and the extended 1M-context tier of Claude Sonnet 4.6 to the catalog. Both routed at the standard 30% discount; cache pricing passes through unchanged.

Prompt caching passthrough verified end-to-end

Verified that `cache_control` blocks survive every hop in the gateway with no semantic difference from direct Anthropic. Cache hit rates observed in production now match direct-Anthropic baselines (60-85% for Claude Code workloads).

OpenAI-compatible /v1/chat/completions endpoint

Anvat now speaks both Anthropic /v1/messages and OpenAI /v1/chat/completions on the same key. Cursor's custom OpenAI endpoint, LangChain ChatOpenAI, and the official OpenAI SDK all work without changes — just swap the base URL.