50% OFF AT LAUNCHDIFFUSION LLM1M CONTEXT1,068 TOKENS/SEC131K MAX OUTPUT
Mercury 3 API Pricing
Mercury 3 is Inception's flagship diffusion LLM, released on October 19, 2026 on the company's own API and through OpenRouter, Baseten and Vercel. The list price is $0.20/M input, $0.02/M cached input and $0.70/M output, and at launch Inception takes 50% off: $0.10 / $0.01 / $0.35 is what an account is charged today. No end date is published, so the tiles hold the list rate and the discount is shown beside it. The context window is 1M tokens, four times Mercury 2.5's, and with this release Mercury 2 and 2.5 closed to new users.
Input - per 1M tokens
$0.20/M
Charged now $0.10/M - 50% off list
Output - per 1M tokens
$0.70/M
Charged now $0.35/M - 50% off list
Cached input - per 1M tokens
$0.02/M
Charged now $0.01/M - 50% off -90%
Effective - agentic blend
$0.104/M
92/8 split - 82% cache
§ 01 / TERMINAL
Run the numbers.
Calculator pre-loaded with Mercury 3's LIST rates, the ones that apply once the launch discount ends. While 50% off runs, halve every result for the bill you get today. Inception has not said when the discount ends.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Real-world presets.
WEB AGENT
Browser navigation step
$0.00236/step
SEARCH AGENT
Query rewrite and answer
$0.00367/query
LONG CONTEXT
Read a whole repository
$0.037/pass
SUPPORT
Ticket reply
$0.00112/ticket
Estimate · inception-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 456
Words 75
Tokens (estimated) 118 tokens
Cost as input · uncached $0.00002 USD
Cost as output · uncached $0.00008 USD
Cost as cached input $0.000002 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Mercury 3 Current | $0.20 cache $0.02 | $0.70 | $0.104 agentic 92/8 | 1M | Latency-bound agents, search and support with a 1M window |
| Mercury Voice | $0.40 cache $0.04 | $1.50 | $0.216 +2.1× · promo | 128K | Voice agents, for enterprise accounts |
| Mercury 2.5 | $0.20 cache $0.02 | $0.75 | $0.108 deprecated · +4% | 260K | The previous generation, closed to new users |
| Mercury Edit 2 | $0.25 cache $0.025 | $0.75 | $0.12 +15% | 32K | Autocomplete and next-edit |
| GPT-6 Luna | $0.10 cache $0.01 | $0.50 | $0.0682 −35% | 1.05M | High-volume work on a closed frontier stack |
| Claude Haiku 5.5 | $0.10 cache $0.01 | $0.50 | $0.0682 −35% | 1M | Anthropic's fast tier with vision |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 +4.6× · promo | 1M | Google's fast multimodal tier |
| DeepSeek V4.1 Flash | $0.30 cache $0.006 | $1.20 | $0.15 +44% | 1M | Cheap 1M-context agent loops with images |
Frequently asked.
What Mercury 3 costs with and without the launch discount, how it compares with the fast tiers it is measured against, and what happens to Mercury 2 and 2.5.
Q · 01 How much does Mercury 3 cost? +
The list price is $0.20/M input, $0.02/M cached input and $0.70/M output, $0.10/M on our 92/8 agentic blend with 82% cache hits. At launch Inception takes 50% off, so an account is charged today
$0.10 / $0.01 / $0.35. Cached input is 90% off fresh input, as on earlier Mercury models. Inception prints the list rates struck through beside the discounted ones on its developer table.Q · 02 When does the 50% discount end? +
Inception has not said. The launch post calls it the price at launch, and neither the developer table nor the models page gives a date. Its last launch discount, 80% on Mercury 2.5, also had no date and is gone from the table by October 21, 2026. That is why this page's tiles and calculator hold the list rate: budget on the list and treat the discount as a saving that can stop without notice.
Q · 03 What changed from Mercury 2.5? +
The window and the output side of the bill. Mercury 3 holds 1M tokens of context against 260K and up to 131,072 output tokens against 65,536, adds an
instant reasoning setting, and lists at $0.70 output against $0.75; input and cached input list at the same $0.20 and $0.02. Inception says it is more than 40 times the size of Mercury 1 and runs at 1,068 tokens per second. Mercury 2.5 and Mercury 2 are closed to new users since this release; accounts that used them before October 19, 2026 keep access until November 15, 2026.Q · 04 Is it cheaper than GPT-6 Luna or Claude Haiku 5.5? +
Not at list. Inception benchmarks Mercury 3 against those two, and both list at
$0.10 input and $0.50 output, so Mercury 3's list card is higher on both. During the launch discount the order changes: $0.10 input ties them and $0.35 output undercuts them. The comparison that favours Mercury is time per task, not dollars per token, and it lasts only as long as the discount does.Q · 05 How fast is it, and does speed change the bill? +
Inception quotes 1,068 tokens per second on NVIDIA GPUs and says that is 4 to 8 times faster than the models it compares against; those are the vendor's figures, not our measurement. Billing is per token as usual, so speed does not lower the token bill. It changes what latency-bound work costs in time: Inception's examples are browser agents that wait on the model at every step, search agents answering inside a few seconds, voice tutoring and customer support.
Q · 06 Is there a free allowance? +
Yes: new API keys come with 100 million free tokens, shared across models. The documentation and the models page now give the same figure; earlier in 2026 the models page said 10 million.
Q · 07 Mercury 3 or Mercury Voice for a voice agent? +
On a self-serve account, Mercury 3: Inception's own voice quickstart and its Vapi and ElevenLabs cookbooks use it. Mercury Voice is tuned for time to first answer token and is sold to enterprise customers through sales, at a list card of
$0.40 input and $1.50 output, above Mercury 3's on both. Its context is 128K against Mercury 3's 1M.