NewModelHubCalculate Your Cost
Mercury 3 pricing is flat: the same per-token rates apply whether your prompt is 1K or 1M tokens. The launch discount has no published end date — switch to Standard pricing to see your post-promotion budget.
Your Estimate
Launch pricing
Cost per request: $0.00
Estimated monthly cost: $0.00
Mercury 3 Rate Card (per 1M tokens)
| Token type | Mercury 3 launch | Mercury 3 standard | Haiku 5.5 ≤100K | Haiku 5.5 >100K |
|---|---|---|---|---|
| Input | $0.10 | $0.20 | $0.10 | $0.50 |
| Cached input / cache read | $0.01 | $0.02 | $0.01 | $0.05 |
| Output | $0.35 | $0.70 | $0.50 | $2.50 |
Mercury 3 rates from Inception's official models page (inceptionlabs.ai/models), verified October 21, 2026. Mercury 3: 1M-token context window, up to 131,072 output tokens per request, model ID mercury-3, OpenAI-compatible API, reasoning effort instant / low / medium / high (default low). Haiku 5.5 rates from Anthropic's official pricing page; its tier is set by total prompt size per request.
What Drives Your Mercury 3 Bill
- Flat rate, no context cliff. Mercury 3 charges the same per-token price at 1K and at 1M prompt tokens. Long-context workloads that would trip Haiku 5.5's 5× tier over 100K tokens cost exactly what the token counts say here.
- Output is the expensive side. Output costs 3.5× the input rate at both standard and launch pricing. Cap max_tokens, and keep answers terse by instruction — output length is the lever that moves the bill.
- Cache everything static. Cached input is priced at 10% of fresh input. System prompts, tool schemas and reference documents you re-send on every call should arrive as cached input tokens, not fresh ones.
- Pin the reasoning effort. Mercury 3 takes an effort setting (instant, low, medium, high) and defaults to low when you omit it. Higher effort burns more output tokens on reasoning — set it deliberately per endpoint instead of inheriting surprises.
- Budget at standard rates. The 50% launch discount has no announced end date. A budget built on $0.10 / $0.35 doubles the day the promotion ends; plan on $0.20 / $0.70 and treat the discount as runway.
- Burn the free tokens first. New API keys come with 100 million free tokens. Prototype, run evals and tune prompts on those before any paid traffic — the free-line estimate above shows how far they stretch at your workload.
FAQ
How much does Mercury 3 cost?
Standard pricing is $0.20 per million input tokens and $0.70 per million output tokens, with cached input at $0.02 per million. A 50% launch discount currently brings that to $0.10 input, $0.35 output and $0.01 cached input per million.
How long does the launch discount last?
Inception has not published an end date for the 50% launch discount. Budget production workloads at the standard rates and treat the launch rates as a temporary bonus.
What are cached input tokens?
Repeated prompt content — system prompts, tool schemas, long static context — served from cache instead of processed fresh. At $0.02 per million standard ($0.01 at launch) they cost one tenth of regular input.
How do the 100 million free tokens work?
Inception's models page states that new API keys come with 100 million free tokens. At roughly 5,000 tokens per chatbot request, that is about 20,000 requests of prototyping and evaluation before you pay anything.
Is Mercury 3 cheaper than Claude Haiku 5.5?
At launch rates Mercury 3 ties Haiku 5.5 on input ($0.10 per million) and beats it on output ($0.35 vs $0.50), with no long-context surcharge across its 1M window — Haiku 5.5 re-prices whole requests at 5× past 100K prompt tokens. At standard rates Mercury 3 input is 2× Haiku 5.5's short-prompt rate, still far below its long-context tier. Run your own workload through both calculators to see which wins for you.
Does the reasoning effort setting change the cost?
Yes. Effort (instant, low, medium, high) defaults to low. Higher effort spends more output tokens on reasoning before the visible answer, and output is Mercury 3's priciest token type — pin the setting explicitly in production.
Methodology & sources: Mercury 3 rates verified against Inception's official models page on October 21, 2026 (inceptionlabs.ai/models); Haiku 5.5 rates against Anthropic's official launch post and pricing page on October 20, 2026. Third-party hosts (OpenRouter, Baseten, Vercel) may price Mercury 3 differently. Estimates exclude taxes and any host markup — your invoice follows metered usage.