What Inception announced
Inception introduced Mercury 3, a diffusion language model that the company says is available through its own API and through named third-party providers. The announcement, dated October 19, 2026 on Inception's blog, describes the model as compatible with the OpenAI API and lists a one million token context window. The company also publishes token pricing, with launch rates of $0.10 input and $0.35 output per million tokens and standard rates of $0.20 and $0.70, and does not state how long the promotional rates will run.
Inception further claims generation speed around 1,068 tokens per second and describes the model as several times faster than comparison products. Secondary coverage, including Jason Hiner's report in The Deep View, repeats the announcement and adds an interview attributed to Inception's chief executive. Those speed and comparison figures remain the vendor's own claims. The captured materials do not show matched serving conditions such as hardware, batch size, concurrency, input lengths or sample sizes, and no independent replication of the quality or speed results was available at the time of writing.
Why the claims need an evaluation, not a switch
A new fast model option is useful precisely because it forces a comparison. A one million token context window does not by itself establish that a system retrieves the right record or answers correctly across an entire conversation. API compatibility does not guarantee that every tool calling, streaming, schema or security behavior a particular deployment depends on will behave identically after a swap. Promotional pricing is a point in time, not a contract, and provider terms can change.
The honest reading of this release is that Inception is offering another candidate for latency sensitive work, with specifications the vendor documents and performance figures the vendor reports. Teams that benefit are the ones that test the candidate against their own representative questions, their own data permissions and their own fallback rules, rather than teams that move production traffic because a headline number looks attractive.
What this means for your business
For a company running customer support, the visible problem is rarely raw token speed alone. Customers wait while staff locate account records, job histories and approved policies, and a fast but unsupported answer can create an expensive error. A concrete system Detailed In Design could build is a customer portal and support workspace that retrieves permissioned business records and approved policies, drafts source linked replies, escalates uncertain cases to staff, and logs every review decision. Within such a system, a model adapter could evaluate Mercury 3 alongside the current model using a fixed set of representative questions, measuring correctness, latency and fallback behavior before any change, with an approved fallback model always available.
Realistic limitations matter as much as the capability. Building this requires approved API access and procurement, a retention and privacy review, permissioned business data, working integrations, and ongoing monitoring. Billing changes, sensitive cases and any dispatch or financial action should stay behind explicit permissions and human approval. This is a proposed custom system from Detailed In Design, not a vendor feature, an existing deployed product, or a partnership with Inception. A reasonable next step is to map one support workflow and test representative questions against your current model before deciding whether a faster option changes the customer experience.
Sources and original reporting
- Introducing Mercury 3 (Inception blog, primary announcement)
- Why Inception's Mercury 3 upends AI economics (The Deep View, Jason Hiner)
- Inception API documentation
Share this story
Use More apps or Copy link for Instagram, Threads, YouTube, and other platforms.