💰 AI Agent Cost Calculator

How the math works

Two billing models drive every total below:

• Per-token billing (API + OpenClaw): messages × tokens × per-token rate. Costs scale linearly with usage; cheap at low volume, can run high at heavy usage.
• Subscription (Cowork, ChatGPT, Claude Code Pro/Max): flat per-seat fee with usage caps. Cheap above a usage threshold; potentially wasted money below it.

Published rates (USD per 1M tokens):

• Anthropic (verified 2026-10-09) — Haiku 5.5 $0.10/$0.50 · Haiku 4.5 $1/$5 · Sonnet 5.5 / Sonnet 5 $2/$10 · Sonnet 4.5 / Sonnet 4.6 $3/$15 · Opus 4.6 / 4.7 / 4.8 / 5 $5/$25 · Opus 5.5 $4/$20 · Fable 5.1 / Mythos 5.1 / Fable 5 $10/$50 (most capable). Opus 5.5 is the current flagship — 1M-token context as both default and maximum, 128k max output, thinking on by default, at $4/$20, down from Opus 5's $5/$25. Haiku 5.5, Opus 4.7, Opus 4.8, Opus 5.5, Opus 5, Fable 5.1, Mythos 5.1, Fable 5 scale output by the effort multiplier (low 1× → max 7×).
• OpenAI (verified 2026-09-24) — GPT-6 Astra $10/$50 · GPT-6 Sol $2/$10 · GPT-6 Luna $0.10/$0.50 · GPT-5.6 Sol $4/$20 · GPT-5.6 Terra $2/$12 · GPT-5.6 Luna $0.20/$1.20 · GPT-5.5 $5/$30 · GPT-5.4 $2.50/$15 · GPT-5.4 mini $0.75/$4.50 · o4-mini $1.10/$4.40 · GPT-5.4-Cyber $12/$60 (not on OpenAI's public price list — unverified). GPT-6 Sol and Luna (September 22) cost half the GPT-5.6 tiers they replace. Long-context requests bill at a higher tier (Sol $4/$15, Astra $20/$75). OpenAI marks GPT-5.6 Sol's $4/$20 as promotional through November 21, 2026.
• Google (verified 2026-09-24) — Gemini 3.8 Flash $0.75/$3.75 (introductory through December 31; $1.50/$7.50 from January 1, 2027) · Gemini 3.1 Pro (Preview) $2/$12, rising to $4/$18 above 200K tokens · Gemini 3.1 Flash-Lite $0.25/$1.50 · Gemini 2.5 Pro $1.25/$10 · Gemini 2.5 Flash $0.30/$2.50 · Gemma 4 local-only (Ollama)
• xAI (verified 2026-09-24) — Grok 4.7 $2/$6 (September 21) · Grok 4.6 $2/$6 · cached input $0.50. At 200K+ tokens the whole request rebills at $4/$12. Grok Bot is billed as a plan allowance, not at these rates — see Grok Bot pricing.
• Meta (verified 2026-09-27) — Muse Spark 1.3 $1.25/$4.25, cached input $0.15, 1M context, via the Meta Model API or OpenRouter. The -contributor variant is $0.10/$0.20 but lets Meta train on your prompts. It's the model behind the Meta Muse agent, whose own plans are a flat $0/$20/$100 a month.
• TypeSafe AI (verified 2026-09-24) — Jev $0.042 input, output free. It returns typed decisions, not text, so compare it only for the classification, routing and gating calls in a workflow — see our Jev guide.
• Open weights via API (OpenRouter list prices, 2026-09-24) — Kimi K3 $3/$15 · Qwen 3.8 2.4T-A95B $2/$6 · GLM-5.3 $0.84/$2.64 · Qwen 3.8 27B $0.42/$3 (Apache-2.0) · MiMo-V2.6 Pro $0.435/$0.87 (MIT, September 21) · DeepSeek V4.1 Flash $0.30/$1.20 at peak, first-party ($0.15/$0.60 off-peak) · GLM-5.3 Flash $0.15/$0.50 (MIT). "Open weights" means the weights are downloadable, not that the licence is permissive — GLM-5.3, Qwen 3.8 2.4T and Kimi K3 use custom licences.
• Open weights local (Ollama + GPU) — Gemma 4 12B, Qwen 3.6 35B MoE: $0/token, roughly $10–18/mo electricity for typical home GPU use. No data leaves your network.
• Subscriptions — Cowork Pro ~$20/user · Business ~$30/user · ChatGPT Plus $20 · Pro $200 · Team $30/seat · OpenClaw self-hosted $0–5/mo

Effort level multiplier (Haiku 5.5, Opus 4.7, Opus 4.8, Opus 5.5, Opus 5, Fable 5.1, Mythos 5.1, Fable 5, GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna, GPT-6 Astra, Gemini 4 Argon, Gemini 3.8 Flash): low 1× · medium 1.3× · high 2× · xhigh 3.5× · max 7×. Multiplier applies to output tokens (where the extra reasoning work shows up).

What this calculator does not include

• Prompt-cache discounts — can cut input cost 50–90% for stable system prompts (see our cost optimization guide)
• Batch API discounts — Anthropic offers 50% off for batch jobs
• Enterprise volume discounts — negotiate above ~$50K/year spend
• Tool-call costs (web search, file uploads) — usually small but vary by platform
• Local GPU electricity if running NemoClaw/Ollama (~$5–20/mo for typical home use)

For most users, real-world spend lands within ±25% of the estimate. The calculator's main job is to spot order-of-magnitude differences between platforms.

Common scenarios