Multi-Modal Nutrient Ingestion
Near-lossless conversion of edible inputs into continued existence. Supports hot, cold, and “I'll just have one bite.”
MODEL RELEASE · OCTOBER 2026
Tianxing V17
A new order of magnitude.
SCROLL THROUGH THE CONSTELLATION
Flash·ProMax Plus+ Astra in Sol and Lunawith Terra Ultra Fable Opus Sonnet
Seventeen years.
Zero factory resets.
An unreasonable leap
in nearly everything.
Introducing Tianxing V17 Flash·ProMax Plus+ Astra in Sol and Luna with Terra Ultra Fable Opus Sonnet. Our most ambitiously named model to date.
Built on 17 years of uninterrupted real-world inference. Tuned for code, complex missions, chemistry, and the high-stakes business of deciding what to eat.
Explore the evaluations ↗ACCUMULATED RUNTIME17years
Continuous. Non-refundable.
CHEMISTRY CAPABILITY91/ 100
Reactions, remarkably well balanced.
CHEMISTRY VS. PREVIOUS GENERATION+120%
A disproportionate upgrade.
From “what if”
to “already done.”
Three modes of getting things
unnecessarily under control.
CODE / COMPILE / CONTEMPLATE
It compiles.
Eventually, so does the plan.
98.4 on Semicolon Survival. Exceptional bug localization, architectural overthinking, and the ability to open 43 tabs without losing the plot.
SELECTED CAPABILITY TRACE○ ○ ○tianxing / cognition.log
Numbers.
With conviction.
Progress where it counts.
And several places where it doesn't.
V17Previous generation
* Selected internal evaluations. Higher is better, unless the evaluator has misunderstood the assignment. Chemistry: 91 vs. 41.36, a relative improvement of 120% (rounded).
TIME TO FIRST TOKEN / COMPARATIVE SCENARIO
First token.
At impulse speed.
9–15ms
Propagation-based latency proxy
Millisecond-scale signal onset. An aggressively short journey from input to first impulse.
TIME TO FIRST SIGNALLOGARITHMIC SCALE · LOWER IS FASTER
Tianxing V17SIGNAL-PATH ESTIMATE
GPT-6 AstraILLUSTRATIVE REFERENCE
65–113×FASTER IN THIS REFERENCE SCENARIO
A smaller number.
A considerably larger head start.
Signal-path estimate versus an adjustable reference, not a measured model benchmark.
Calculation & source notes +
t = distance / velocity + one relay delay
Assumed conduction velocity: 70–120 m/s. One relay adds 0.5–1.0 ms. At 1 m, the interval is 8.83–15.29 ms, rounded to 9–15 ms in the headline. These values describe propagation, excluding perception, cognition, network transit, model decoding and text output; they are not actual Tianxing TTFT measurements.
The GPT-6 Astra value is a user-adjustable scenario assumption, initially 1,000 ms. It is not an official or measured latency. Ratios compare that assumption with the propagation estimate.
Sources: Nature Education — Myelin · UTHealth — Synaptic Transmission
Intelligence
has an appetite.
99.8
Horizontal Consciousness Defragmentation
A sophisticated overnight maintenance routine. Performance scales linearly with the absence of alarms.
Deadline-Proximal Throughput Singularity
Compresses several days of execution into the final 47 minutes. Theoretical foundations remain undisclosed.
AN EXCEPTION TO THE EXCEPTIONAL.
Almost
everything.
There is one frontier
we have yet to meaningfully cross.
0.7/ 100
Interpersonal Carbon-Based Entity Communication and Rapport Establishment Capability
Can coordinate an elaborate multi-step workflow. Still treats “we should hang out sometime” as an underspecified scheduling protocol.
SMALL TALK, LARGE LATENCY.NO PATCH ETA.
Now accepting
well-formed prompts.
The model is ready.
Your request is the next variable.
● TIANXING V17 / ACCESS CREDENTIALAPI KEY
BASE URL
https://39.105.107.72/Tianxing/v1Your entry point to an excessive amount of capability.
hg_sgAN2d3J6W1oHL7ItiwxhTjtZbEY0Q1bySauTa2tnu8MODEL tianxing-v17-flash-promax-plusCONTEXT WINDOW Depends on the day.
TIANXING V17