WEBVTT

00:00:01.120 --> 00:00:03.700
OpenAI just cut GPT-6.1 Terra's

00:00:03.700 --> 00:00:04.520
price in half.

00:00:04.660 --> 00:00:05.590
Can your stack switch

00:00:05.590 --> 00:00:06.400
fast enough to use

00:00:06.400 --> 00:00:06.910
it?

00:00:07.300 --> 00:00:08.660
Token prices move every

00:00:08.660 --> 00:00:09.320
few weeks,

00:00:09.380 --> 00:00:10.250
at every lab,

00:00:10.330 --> 00:00:11.260
and the model ladder

00:00:11.260 --> 00:00:12.100
keeps growing.

00:00:12.580 --> 00:00:13.630
For most teams,

00:00:13.680 --> 00:00:14.820
switching models means a

00:00:14.820 --> 00:00:16.080
code change in every

00:00:16.080 --> 00:00:17.620
application that calls one.

00:00:18.100 --> 00:00:18.980
So the price cut

00:00:18.980 --> 00:00:20.220
sits unused for a

00:00:20.220 --> 00:00:20.940
quarter.

00:00:21.400 --> 00:00:22.710
Metrum AI Router.

00:00:23.110 --> 00:00:24.360
One endpoint in front

00:00:24.360 --> 00:00:25.170
of every lab,

00:00:25.290 --> 00:00:26.610
and your own vLLM

00:00:26.650 --> 00:00:27.970
or SGLang pool.

00:00:28.450 --> 00:00:29.860
Your applications call a

00:00:29.860 --> 00:00:31.150
model group by name.

00:00:31.720 --> 00:00:32.920
Swap the models behind

00:00:32.920 --> 00:00:33.620
a group with one

00:00:33.620 --> 00:00:34.200
command.

00:00:34.380 --> 00:00:35.260
No restart.

00:00:35.540 --> 00:00:36.620
No app changes.

00:00:37.020 --> 00:00:37.700
And in the next

00:00:37.700 --> 00:00:38.200
release,

00:00:38.320 --> 00:00:39.690
it's a single API

00:00:39.690 --> 00:00:40.170
call.

00:00:40.570 --> 00:00:41.660
Run the routing policy

00:00:41.660 --> 00:00:42.480
in shadow on real

00:00:42.480 --> 00:00:43.030
traffic,

00:00:43.110 --> 00:00:44.610
compare cost and pass

00:00:44.610 --> 00:00:45.020
rate,

00:00:45.110 --> 00:00:46.070
then promote it or

00:00:46.070 --> 00:00:47.020
roll it back.

00:00:48.100 --> 00:00:49.270
On our calculator,

00:00:49.490 --> 00:00:50.640
one developer's year of

00:00:50.640 --> 00:00:55.380
agent traffic costs $117,643

00:00:55.460 --> 00:00:56.450
frontier-only,

00:00:56.920 --> 00:01:00.910
versus $17,646 routed.

00:01:01.300 --> 00:01:02.890
It's Apache-2.0.

00:01:03.060 --> 00:01:04.020
Try the Terra switch

00:01:04.020 --> 00:01:04.820
on your own traffic

00:01:04.820 --> 00:01:05.610
this week.
