Claude API pricing — list price, then 30% off.
Every Claude model on the Anthropic API. Per-million token rates, 5-minute and 1-hour cache writes, cache hits, deprecated and retired tiers — everything you'd see on Anthropic's pricing page, plus the Anvat effective rate after our flat 30% discount.
Same wire format
Point ANTHROPIC_BASE_URL at api.anvat.app. SDK, tool use, streaming, MCP, vision — all passthrough.
-30% on every bucket
Input, output, 5m / 1h cache write, cache hits, reasoning. No bucket excluded, no rounding.
Bill matches the table
Every dashboard line item references the rate below. We show our math; you can verify any request.
Opus
Flagship reasoning tier
| Model | Context | Input | 5m cache write | 1h cache write | Cache hits | Output | Anvat in / out (-30%) |
|---|---|---|---|---|---|---|---|
Claude Opus 4.8 claude-opus-4-8 | 200K | $5 | $6.25 | $10 | $0.50 | $25 | $3.50 / $17.50 |
Claude Opus 4.7 claude-opus-4-7 | 200K | $5 | $6.25 | $10 | $0.50 | $25 | $3.50 / $17.50 |
Claude Opus 4.6 claude-opus-4-6 | 200K | $5 | $6.25 | $10 | $0.50 | $25 | $3.50 / $17.50 |
Claude Opus 4.5 claude-opus-4-5 | 200K | $5 | $6.25 | $10 | $0.50 | $25 | $3.50 / $17.50 |
Claude Opus 4.1deprecated claude-opus-4-1 | 200K | $15 | $18.75 | $30 | $1.50 | $75 | $10.50 / $52.50 |
Claude Opus 4deprecated claude-opus-4 | 200K | $15 | $18.75 | $30 | $1.50 | $75 | $10.50 / $52.50 |
Sonnet
Workhorse — best price / performance
| Model | Context | Input | 5m cache write | 1h cache write | Cache hits | Output | Anvat in / out (-30%) |
|---|---|---|---|---|---|---|---|
Claude Sonnet 4.6 claude-sonnet-4-6 | 200K 1M-token tier available at 2× the base rate. | $3 | $3.75 | $6 | $0.30 | $15 | $2.10 / $10.50 |
Claude Sonnet 4.5 claude-sonnet-4-5 | 200K | $3 | $3.75 | $6 | $0.30 | $15 | $2.10 / $10.50 |
Claude Sonnet 4deprecated claude-sonnet-4 | 200K | $3 | $3.75 | $6 | $0.30 | $15 | $2.10 / $10.50 |
Sonnet's 1M-token context tier bills at 2× the base rate (input + output + cache writes all double). Use the standard 200K tier unless you genuinely need the extended window — it's half the price and rarely the bottleneck.
Haiku
Cheap, fast, surprisingly capable
| Model | Context | Input | 5m cache write | 1h cache write | Cache hits | Output | Anvat in / out (-30%) |
|---|---|---|---|---|---|---|---|
Claude Haiku 4.5 claude-haiku-4-5 | 200K | $1 | $1.25 | $2 | $0.10 | $5 | $0.70 / $3.50 |
Claude Haiku 3.5retired claude-3-5-haiku Bedrock & Vertex AI only | 200K | $0.80 | $1 | $1.60 | $0.08 | $4 | $0.56 / $2.80 |
Prompt-caching math (across every current model)
Anthropic uses the same multipliers across every current Claude tier. Once you know the input rate, the cache math is mechanical:
| Cache hits & refreshes | 0.10 × input | 90% off list once a chunk is cached |
|---|---|---|
| Cache write — 5-minute TTL | 1.25 × input | Default for tight agent loops |
| Cache write — 1-hour TTL | 2.00 × input | Pays off for long sessions and audits |
Tip: a Claude Code session that hits 70-85% cache rate cuts your effective input cost by 60-80% — bigger than any discount we can offer. Stack caching with the Anvat -30%, and a typical month lands at roughly 40-50% of what Anthropic-direct would bill.
What -30% looks like for a typical day
Some concrete numbers from production Claude Code traffic flowing through Anvat in May 2026 — same prompts, same models, just different list rate.
| Workload | Anthropic direct | Anvat (-30%) | Saved |
|---|---|---|---|
| Solo Claude Code — 4-6h/day, Sonnet 4.6 + caching | $120 | $84 | −$36 |
| 10-engineer team on Claude Code (Sonnet 4.6) | $1,200 | $840 | −$360 |
| Cursor power user — heavy Composer (Sonnet 4.6) | $220 | $154 | −$66 |
| Coding-agent SaaS backend on Opus 4.8 (50K MAU) | $4,800 | $3,360 | −$1,440 |
| RAG knowledge-base Q&A with cached system prompt | $680 | $476 | −$204 |
Direct column = list-price Anthropic billing for the same token mix. Anvat column = same usage at -30%. Numbers are monthly USD; your mileage varies with cache hit rate and routing.
Verify these numbers yourself
- 1. Open anthropic.com/pricing in another tab. Every list-price column above matches it.
- 2. Multiply any list rate by
0.70. That's exactly what the Anvat column shows. - 3. Open your request log after any API call — the per-bucket breakdown lines up with the table to 5-6 decimals of USD.
Same Anthropic models. 30% less.
Point your Anthropic SDK at api.anvat.app/v1 and the rates above are what you'll pay — no markup, no rounding, no hidden fees.