# AI model pricing

Current $/million-token pricing for every popular LLM, timestamped and sourced. Refreshed daily where a provider's page can be parsed safely (Anthropic, OpenAI); every other provider is checked by hand and its "verified" date ages visibly rather than being guessed.

Interactive version: https://www.reelier.com/models. Cost calculator (plug in your own token counts): https://www.reelier.com/tools/agent-cost-calculator

## Anthropic

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| Claude Fable 5 | $10.00 | $50.00 | 2026-07-22 | https://platform.claude.com/docs/en/about-claude/pricing |
| Claude Opus 4.8 | $5.00 | $25.00 | 2026-07-22 | https://platform.claude.com/docs/en/about-claude/pricing |
| Claude Sonnet 5 | $2.00 | $10.00 | 2026-07-22 | https://platform.claude.com/docs/en/about-claude/pricing |
| Claude Haiku 4.5 | $1.00 | $5.00 | 2026-07-22 | https://platform.claude.com/docs/en/about-claude/pricing |

**Claude Fable 5:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.19 — a Reelier replay of the same workflow: $0.00
**Claude Opus 4.8:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.10 — a Reelier replay of the same workflow: $0.00
**Claude Sonnet 5:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.04 — a Reelier replay of the same workflow: $0.00 (Introductory pricing in effect through 2026-08-31; standard pricing ($3/$15) takes over 2026-09-01.)
**Claude Haiku 4.5:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.02 — a Reelier replay of the same workflow: $0.00

## OpenAI

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| GPT-5.6 Sol | $5.00 | $30.00 | 2026-07-22 | https://developers.openai.com/api/docs/pricing |
| GPT-5.6 Terra | $2.50 | $15.00 | 2026-07-22 | https://developers.openai.com/api/docs/pricing |
| GPT-5.6 Luna | $1.00 | $6.00 | 2026-07-22 | https://developers.openai.com/api/docs/pricing |

**GPT-5.6 Sol:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.10 — a Reelier replay of the same workflow: $0.00
**GPT-5.6 Terra:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.05 — a Reelier replay of the same workflow: $0.00
**GPT-5.6 Luna:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.02 — a Reelier replay of the same workflow: $0.00

## Google

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| Gemini 3.1 Pro | $2.00 | $12.00 | 2026-07-22 | https://ai.google.dev/gemini-api/docs/pricing |
| Gemini 3.6 Flash | $1.50 | $7.50 | 2026-07-22 | https://ai.google.dev/gemini-api/docs/pricing |
| Gemini 3.5 Flash-Lite | $0.3000 | $2.50 | 2026-07-22 | https://ai.google.dev/gemini-api/docs/pricing |

**Gemini 3.1 Pro:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.04 — a Reelier replay of the same workflow: $0.00 (Tiered by prompt size; rate shown is the <=200k input-token tier (>200k is $4/$18).)
**Gemini 3.6 Flash:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.03 — a Reelier replay of the same workflow: $0.00
**Gemini 3.5 Flash-Lite:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.01 — a Reelier replay of the same workflow: $0.00

## xAI

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| Grok 4.5 | $2.00 | $6.00 | 2026-07-22 | https://docs.x.ai/docs/models |

**Grok 4.5:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.04 — a Reelier replay of the same workflow: $0.00 (Tiered by prompt size; rate shown is the <200k input-token tier (>=200k is $4/$12).)

## Moonshot (Kimi)

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| Kimi K3 | $3.00 | $15.00 | 2026-07-22 | https://platform.kimi.ai/docs/pricing/chat-k3 |
| Kimi K2.6 | $0.9500 | $4.00 | 2026-07-22 | https://platform.kimi.ai/docs/pricing/chat-k26 |

**Kimi K3:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.06 — a Reelier replay of the same workflow: $0.00 (Cache-miss input rate shown; cache-hit input is $0.30/Mtok.)
**Kimi K2.6:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.02 — a Reelier replay of the same workflow: $0.00 (Cache-miss input rate shown; cache-hit input is $0.16/Mtok.)

## DeepSeek

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| DeepSeek V4 Pro | $0.4350 | $0.8700 | 2026-07-22 | https://api-docs.deepseek.com/quick_start/pricing |
| DeepSeek V4 Flash | $0.1400 | $0.2800 | 2026-07-22 | https://api-docs.deepseek.com/quick_start/pricing |

**DeepSeek V4 Pro:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.01 — a Reelier replay of the same workflow: $0.00 (Cache-miss input rate shown; cache-hit input is $0.003625/Mtok.)
**DeepSeek V4 Flash:** a typical agent workflow (~18k tokens, the README benchmark) ≈ < $0.01 — a Reelier replay of the same workflow: $0.00 (Cache-miss input rate shown; cache-hit input is $0.0028/Mtok.)

## Mistral

| Model | Input $/Mtok | Output $/Mtok | Verified | Source |
| --- | --- | --- | --- | --- |
| Mistral Large 3 | $0.5000 | $1.50 | 2026-07-22 | https://mistral.ai/pricing/api/ |
| Mistral Medium 3.5 | $1.50 | $7.50 | 2026-07-22 | https://mistral.ai/pricing/api/ |
| Mistral Small 4 | $0.1500 | $0.6000 | 2026-07-22 | https://mistral.ai/pricing/api/ |

**Mistral Large 3:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.01 — a Reelier replay of the same workflow: $0.00
**Mistral Medium 3.5:** a typical agent workflow (~18k tokens, the README benchmark) ≈ $0.03 — a Reelier replay of the same workflow: $0.00
**Mistral Small 4:** a typical agent workflow (~18k tokens, the README benchmark) ≈ < $0.01 — a Reelier replay of the same workflow: $0.00

## Method

Every row is checked against the provider's own pricing page (Source column above). A daily cron re-fetches pages with a reliable, defensively-parsed per-provider extractor and applies a guardrail: a parse failure, or a parsed price more than 50% away from the stored one, leaves the number untouched and marks the row stale rather than writing a guess. Providers without a reliable extractor are checked by hand; their verified date simply ages until the next manual check.

The per-model cost above is computed against one fixed assumption: a benchmark agent run of 18,390 input tokens + 136 output tokens — the measured average from a real npm-registry agent workflow (N=10), the same number https://www.reelier.com/tools/agent-cost-calculator uses. The $0.00 replay column is not a discount: Level-0 replay re-executes the recorded tool calls and never constructs an LLM client, so it spends 0 tokens by construction, verified from the run record on every replay (llmInputTokens === 0). See https://www.reelier.com/learn/level-0-replay

Install: `npm i -g reelier && reelier init` — record the run once, replay it on a schedule, keep the receipt.

Skills: https://www.reelier.com/skills · Cost calculator: https://www.reelier.com/tools/agent-cost-calculator