LLM API token pricing reference
Silicon Analysts tracks LLM API list prices in US dollars per million tokens across Anthropic (Claude), OpenAI (GPT), Google (Gemini), xAI (Grok), DeepSeek, and open-source hosting (Llama on Together AI). Each row records the standard pay-as-you-go input and output token price and, where published, the cached-input price, citing the provider's own pricing page. A weekly agent cross-checks public sources and a daily append-only ledger accrues the price history, so repricings — introductory windows ending, price cuts, new model generations — are captured as dated movements.
LLM API Pricing — $ per Million Tokens by Model (2026)
As of Jul 2026, frontier-model list prices span $10.00 per million input tokens (Claude Fable 5) down to fractions of a dollar for open-weight models — with output tokens typically 3–6x the input price. The model layer joins the stack: the same daily ledger that tracks wafer, chip, and GPU-rental costs now tracks what a token costs.
Last updated: Jul 2026 ·
Current list prices (USD per million tokens)
| Provider | Model | Input | Output |
|---|---|---|---|
| Anthropic | Claude Fable 5 | $10.00 | $50.00 |
| Anthropic | Claude Opus 4.8 | $5.00 | $25.00 |
| Anthropic | Claude Sonnet 5 | $2.00 | $10.00 |
| Anthropic | Claude Haiku 4.5 | $1.00 | $5.00 |
| OpenAI | GPT-5.6 Sol | $5.00 | $30.00 |
| OpenAI | GPT-5.6 Luna | $1.00 | $6.00 |
| Gemini 3.1 Pro Preview | $2.00 | $12.00 | |
| Gemini 3.5 Flash | $1.50 | $9.00 | |
| xAI | Grok 4.5 | $2.00 | $6.00 |
| DeepSeek | DeepSeek V4 Pro | $0.43 | $0.87 |
| DeepSeek | DeepSeek V4 Flash | $0.14 | $0.28 |
| Together | Llama 4 Maverick (Together) | $0.27 | $0.85 |
Standard pay-as-you-go list prices; batch/flex tiers and negotiated rates excluded. Claude Sonnet 5 shows introductory pricing (in effect through Aug 31, 2026). Tiered prices (Gemini) use the base context tier.
Want this table as CSV/JSON? Request it →
Claude Fable 5 — $/Mtok, daily ledger
Daily ledger; the llm_pricing domain begins accruing July 2026 (per-domain start may be later than the 2026-06-06 ledger start). Append-only — cannot be backfilled.
GPT-5.6 Sol — $/Mtok, daily ledger
Cross-source dispersion (consensus low/high per price) is recorded in the ledger from day one and will surface on these charts as the series matures.
Methodology
Prices are standard pay-as-you-go LIST prices in USD per million tokens, taken from each provider's own public pricing page (linked per row). Batch, flex, and priority tiers, fine-tuning rates, and negotiated/enterprise pricing are excluded. Where a provider tiers prices by context length (Google Gemini), the base tier is shown. Introductory list pricing that is currently in effect (Claude Sonnet 5 through Aug 31, 2026) counts as the current price — the scheduled step-up will be captured by the ledger as a dated movement.
A weekly refresh agent (Tuesdays 12:00 UTC) searches public sources for each tracked model, reconciles readings across independent hosts (median consensus, dispersion band), and publishes review-by-exception: values corroborated by the provider's own domain or two-plus independent sources update automatically; single-source readings and moves larger than 50% are held for human review. The daily snapshot then freezes the table into an append-only ledger, so the series cannot be backfilled or rewritten.
For full data provenance and terms of use, see Terms & Data Provenance.
See what the silicon under these tokens costs →
AI Accelerator Manufacturing Costs — H100, B200, GB200 full BOM breakdown including wafer, HBM, and packaging.
See what changed recently →
What Changed — recent movements in LLM API pricing, GPU rental, wafer pricing, and chip cost, tracked daily with provenance.