AI API Pricing
The normalized observation layer keeps endpoint identity, price fields, region, effective time, source and review state together.
See normalization rulesPublic research beta · Official index withheld
Compare input, cached-input, output and batch rates for ten model endpoints across four providers, observed on 24 August 2026. Check the linked provider source before budgeting; prices and access may have changed. For workload estimates, the separate token calculator uses unexpired starter rates or your own rates; it does not load this stale KAPI snapshot.
Release kapi-shadow-2026-08-24 · observed 24 August 2026 · independent recomputation passed
The prices are public; the index is not. These dated official-source observations are retained for historical inspection, not current quotes. Kingy is not publishing an official KAPI headline, ranking or cheapest-model claim until thirteen consecutive weekly baselines and a separate publication authorization pass. A missing price means “not published,” never zero.
Normalized price observations
USD per one million tokens unless marked otherwise. “Eligible” refers to the governed future index cohort. These historical rows are not available as current-price presets in the separate token calculator.
| Model endpoint | Provider | Input | Cached input | Output | Batch in / out | Index status | US evidence | Evidence & tools |
|---|---|---|---|---|---|---|---|---|
| GPT-5.4 minigpt-5.4-mini | OpenAI | $0.75 | $0.075 | $4.5 | $0.375 / $2.25 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
| GPT-5.4gpt-5.4 | OpenAI | $2.5 | $0.25 | $15 | $1.25 / $7.5 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
| GPT-5.5gpt-5.5 | OpenAI | $5 | $0.5 | $30 | $2.5 / $15 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
| Claude Haiku 4.5claude-haiku-4-5-20251001 | Anthropic | $1 | $0.1 | $5 | $0.5 / $2.5 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
| Claude Sonnet 4.6claude-sonnet-4-6 | Anthropic | $3 | $0.3 | $15 | $1.5 / $7.5 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
| Claude Opus 4.8claude-opus-4-8 | Anthropic | $5 | $0.5 | $25 | $2.5 / $12.5 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
| DeepSeek V4 Flashdeepseek-v4-flash | DeepSeek | $0.44Conservative peak rate; UTC bands disclosed | $0.014 | $1.32 | Not published | Not eligibleUS_AVAILABILITY_UNVERIFIED | UNVERIFIED_US | Official source Model profile Open separate calculator |
| DeepSeek V4 Prodeepseek-v4-pro | DeepSeek | $1.32Conservative peak rate; UTC bands disclosed | $0.044 | $3.96 | Not published | Not eligibleUS_AVAILABILITY_UNVERIFIED | UNVERIFIED_US | Official source Model profile Open separate calculator |
| Grok Build 0.1grok-build-0.1 | xAI | $1 | $0.2 | $2 | Not published | EligibleBATCH_RATE_WITHHELD_OFFICIAL_SURFACE_CONFLICT | VERIFIED_US | Official source Model profile Open separate calculator |
| Grok 4.3grok-4.3 | xAI | $1.25 | $0.2 | $2.5 | $1 / $2 | EligibleNo exclusion reason | VERIFIED_US | Official source Model profile Open separate calculator |
Long-context multipliers, prompt-cache write costs and provider-specific conditions are preserved in the public data and calculator disclosures. Prices are evidence, not a model-quality verdict.
One data layer, three useful surfaces
The normalized observation layer keeps endpoint identity, price fields, region, effective time, source and review state together.
See normalization rulesThe same observation appears as source-checked price evidence on each matching canonical model record.
Browse model profilesThe separate token calculator uses unexpired starter rates or your custom rates, then adds cache mix, retries, accepted-result rate and review labour. KAPI-backed current pricing remains paused.
Use the separate token calculatorTrust contract
Each row cites an official provider surface and retains a checksum of the captured evidence.
Unsupported batch, cache or regional facts remain null and cannot silently enter a calculation.
A usable official USD price can coexist with an index exclusion when regional evidence is incomplete.
The official headline, rankings and shadow computation remain unpublished during this research beta.
The sealed release passed a separate calculation path before any public integration was prepared.
This table retains release kapi-shadow-2026-08-24. The separate token calculator has its own rate checks. Neither its starter rates nor the model pages make this historical table current.
Official evidence catalog
Provider pricing, model identity and availability evidence used by this release.
47d5f784e39e2ae1…4d50758e118b31e5…18890b9879019742…263771821bdd1d9b…523f4139b805a4b7…419acbc15dfff24f…be6ae8d0be5b5ee4…d321546b99bc7706…fdb9e346fb09aef7…55c9c6849d999254…645c7bef06e33a9e…For researchers and builders
The public release exposes observations, source metadata, eligibility and withheld-index status. It excludes the internal shadow headline, creator minima and computed ranked results.