llmburn.

← All models

Anthropic: Claude Sonnet 4

Last Verified 2026-10-01 ✓ 3 independent sources agree context 200,000 JSON ↓

Pricing matrix USD per 1M tokens

FieldConsensusEvidence
Input $3 ✓ 3 independent sources 3 sources
Output $15 ✓ 3 independent sources 3 sources
Cache read $0.3 ✓ 3 independent sources 3 sources
Cache write (5m) $3.75 ✓ 3 independent sources 3 sources
Cache write (1h) $6 single source 2 sources
Reasoning tokens Unverified(未验证,不填 0) not published
证据链:各源原值并列(audit trail,点击展开)
FieldSourceRaw value (USD/1M)IndependentObserved
input openrouter 3.0000 yes 2026-10-01 08:00
input litellm 3.0000 yes 2026-10-01 08:00
input litellm <span class="badge mute">mirror</span> 3.0000 no 2026-10-01 08:00
input modelsdev 3.0000 yes 2026-10-01 08:00
input modelsdev 3.0000 yes 2026-10-01 08:00
input modelsdev 3.0000 yes 2026-10-01 08:00
input modelsdev 3.0000 yes 2026-10-01 08:00
output openrouter 15.0000 yes 2026-10-01 08:00
output litellm 15.0000 yes 2026-10-01 08:00
output litellm <span class="badge mute">mirror</span> 15.0000 no 2026-10-01 08:00
output modelsdev 15.0000 yes 2026-10-01 08:00
output modelsdev 15.0000 yes 2026-10-01 08:00
output modelsdev 15.0000 yes 2026-10-01 08:00
output modelsdev 15.0000 yes 2026-10-01 08:00
cache_read openrouter 0.3000 yes 2026-10-01 08:00
cache_read litellm 0.3000 yes 2026-10-01 08:00
cache_read litellm <span class="badge mute">mirror</span> 0.3000 no 2026-10-01 08:00
cache_read modelsdev 0.3000 yes 2026-10-01 08:00
cache_read modelsdev 0.3000 yes 2026-10-01 08:00
cache_read modelsdev 0.3000 yes 2026-10-01 08:00
cache_read modelsdev 0.3000 yes 2026-10-01 08:00
cache_write openrouter 3.7500 yes 2026-10-01 08:00
cache_write litellm 3.7500 yes 2026-10-01 08:00
cache_write litellm <span class="badge mute">mirror</span> 3.7500 no 2026-10-01 08:00
cache_write modelsdev 3.7500 yes 2026-10-01 08:00
cache_write modelsdev 3.7500 yes 2026-10-01 08:00
cache_write modelsdev 3.7500 yes 2026-10-01 08:00
cache_write modelsdev 3.7500 yes 2026-10-01 08:00
cache_write_1h openrouter 6.0000 yes 2026-10-01 08:00
cache_write_1h litellm <span class="badge mute">mirror</span> 6.0000 no 2026-10-01 08:00

Long-context tier jumps ⚠

When prompt tokens cross a threshold, the entire request is re-priced at the higher tier — not just the excess. This is the #1 hidden cost on long-context workloads.

$0.68220K tokens200K跨过 200,000 tokens:$0.6150 / 请求(整请求重算)
Prompt ≥ tokensinputoutputcache read
0 (base)$3$15$0.3
200,000未验证未验证未验证

Run this model through the real-bill calculator →

Sources

This is an estimate based on public sticker prices, not your invoice. Machine-readable: /data/v1/pricing/anthropic-claude-sonnet-4.json