Run AI-assisted security audits on Opus 4.8 at 30% off list

Frontier-model access is a base-rate security investment in 2026

AI-assisted security research moved from demo to discipline this year. Independent researcher Taylor Hornby used Claude Opus 4.8 to discover a critical soundness bug in Zcash's Orchard pool that had been live since 2022. Project Glasswing partners surfaced 23,000+ candidate vulnerabilities across hyperscaler codebases in eight weeks. The capability is real — but direct-Anthropic pricing at $15/$75 per MTok makes large-scale auditing expensive. Heavy prompt caching combined with the standard 30% gateway discount can cut the bill roughly 70-80%.

Point your audit tooling at api.anvat.app

Anvat is an Anthropic-compatible gateway. Your existing Anthropic SDK code, audit scripts, GitHub Actions, and IDE plugins all work unchanged — just swap the base URL and use an Anvat key. The wire format is identical, prompt caching passes through, tool use works, and you get full activity logging with per-request cost attribution. Most security teams see Opus 4.8 token spend drop 70-80% within the first month after switching, thanks to the combination of the 30% gateway discount and the high cache-hit rates that audit workflows naturally produce.

Why this beats the obvious alternative

• 30% off Opus 4.8 list price

  $15/$75 → $10.50/$52.50 effective. Direct discount with no commitments.

• Prompt caching for repeated audit context

  Spec + implementation cached at 10% of input rate — audit workflows hit 70-85% cache rate naturally.

• Same wire format as direct Anthropic

  Your existing audit scripts, GitHub Actions, IDE plugins work unchanged.

• Per-request cost ledger

  Dashboard surfaces exact cost per audit run — useful for chargeback or grant-funded research.

• $2 free credit on signup

  Enough to run a first-pass audit on a small codebase before committing.

Quickstart

Spec-vs-implementation diff audit (the pattern that found the Zcash bug)

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://api.anvat.app/v1",
  authToken: process.env.ANVAT_API_KEY,
});

const review = await client.messages.create({
  model: "claude-opus-4-8",
  max_tokens: 4096,
  system: [
    {
      type: "text",
      text: SECURITY_AUDIT_SYSTEM_PROMPT,
      cache_control: { type: "ephemeral" },
    },
    {
      type: "text",
      text: protocolSpec, // RFC / paper / written spec
      cache_control: { type: "ephemeral" },
    },
    {
      type: "text",
      text: implementationCode,
      cache_control: { type: "ephemeral" },
    },
  ],
  messages: [{
    role: "user",
    content: `For each invariant the spec requires, walk through how
the implementation enforces it. Flag any check that is loose,
ambiguous, or absent.`,
  }],
});

FAQ

Can I use the same model that found the Zcash bug?

Yes. The Zcash discovery used Claude Opus 4.8, which is the same publicly available flagship Anvat routes by default. Anvat passes the request through unchanged — same model weights, same inference endpoint, 30% off list price.

Can I get access to Claude Mythos through Anvat?

No — Mythos Preview is gated behind Anthropic's Project Glasswing program for vetted critical-infrastructure organizations only. Anvat exposes the publicly available Anthropic catalog (Opus 4.8, Sonnet 4.6, Haiku 4.5). When Anthropic ships a generally available Mythos-class model, we will support it day-of via the same wire format.

Does prompt caching work for audit workflows?

Yes — and it's the single highest-impact optimisation. Audit workflows naturally re-ask similar questions across the same spec + implementation prefix. Caching the prefix drops input cost to roughly 10% on cached portions. Combined with Anvat's 30% gateway discount, this lands at around 7% of direct-Anthropic input cost for the cached parts of the prompt.

Will my exploits / findings be stored anywhere?

Anvat logs only billing metadata — model, token counts, latency, status, cost. Prompt and completion bodies are never persisted by the gateway. For audit work involving sensitive material, this matches the operational expectations of most responsible-disclosure workflows.

What's the per-audit cost in practice?

A representative protocol audit (10K-token spec + 50K-token implementation, 200 turns over a session) costs roughly $2-5 on direct Anthropic Opus 4.8, $1.40-3.50 via Anvat with caching. For comparison, the equivalent in human auditor hours is typically $5K-50K.