Inception · llm model
Mercury 3
Mercury 3 is Inception's latest language model in our dataset, released Oct 19, 2026. Proprietary, API access only. Inception's diffusion-based language model with a 1M-token context, launched Oct 19, 2026 and billed at launch for half its list price. Verified against vendor sources on Oct 26, 2026.
Reported Last verified · All models · JSON · Markdown
- ReleasedOct 19, 2026
- Context window1M1,000,000 tokens
- LicenceProprietaryProprietary (API)
Modalities not verified
Key facts
| Fact | Value | Source |
|---|---|---|
| Vendor | Inception | n/a |
| Released | Oct 19, 2026 Blog post dated Oct 19, 2026; early-access partners had it earlier. | source |
| Status | generally available | n/a |
| Context window | 1,000,000 tokens (1M) Stated as 1M tokens, up from 260K for Mercury 2.5. | source |
| Licence | Proprietary (API) Available through the Inception API Platform, OpenRouter, Baseten and Vercel; no weights are offered. | source |
| Pricing | $0.1 in / $0.35 out per 1M tokens Launch price, 50% off the list price of $0.20 input and $0.70 output. Cached tokens are 90% off the input price. | source |
Where Mercury 3 sits in our data
- Released Oct 19, 2026 (week 43 of 2026), in October 2026. Other models in our data released that month: GPT-6.1 Terra, Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, Nemotron 3.5 Bolt E4B (4GB), Voxtral 3, MiniMax M3.1 Flash.
- Of the 60 models in our data with a published context window, Mercury 3 has the number 20 largest (1M tokens). Our data is not every model on the market.
- All releases by date: AI model releases by month · new AI models this week. Choosing for code: best LLM for coding.
What the sources say
These are claims from the linked pages, not our measurements.
- Inception says Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 on its benchmarks while running 4-8x faster at 1,068 tokens per second on NVIDIA GPUs. These are vendor claims.
What we tested
Nothing yet. We have not run Mercury 3 ourselves, so every fact on this page is Reported from the sources below. When we do test it, the scope will be stated here and in our test logs. See methodology.
Not verified, left out
We could not confirm these on a vendor page, so we do not state them.
- Parameter count was not disclosed.
- Launch pricing has no stated end date.
- OpenRouter's listing shows Oct 16, 2026 (early access); we use Inception's own Oct 19 launch date.