kwindla - Introducing PhoneLLM, an open model for voice agents. GPT 5.6 Terra

01 · Hook · 0:00–0:00

A split screen shows a man in a black t-shirt resting his chin on his hand, gazing at a screen on the left, while a glowing blue and purple orb holding a row of small white squares pulses on the right against black.

Techniques: Split screen

Camera: Static

02 · Hook · 0:00–0:12

Split screen: a man in a black t-shirt sits at a desk with a tablet, talking, on the left; on the right a glowing blue-purple orb pulses on black, its center icon morphing between dots and vertical bars.

Techniques: Split screen, Morph

Camera: Static

03 · Product intro · 0:12–0:14

A black frame holds a left-aligned title block. The small purple kicker and the larger white headline fade in from dim to full brightness, then hold on a locked-off frame.

On-screen text: “PHONEBENCH • RESULTS”, “PhoneBench Alpha 1”

Camera: Static

04 · Product intro · 0:14–0:16

A dark benchmark results table fills in row by row, score bars extending and figures updating as each model's data appears. The Pipercat PhoneLLM 30B Alpha 1 row is highlighted with a purple outline. A dim portrait of a person fades in along the right edge.

Camera: Static

05 · Product intro · 0:16–0:25

Split layout: a dark results table titled PhoneBench Alpha 1 fills the left, rows of model names with blue score bars, the Pipewheel PhoneLLM 30B Alpha 1 row lit in purple; a presenter in a black tee talks on the right. A purple frame draws around the score column and a small bullseye icon pops in beside it as he gestures.

Techniques: Split screen

Camera: Static

06 · Product intro · 0:25–0:30

A dark model-results table fills the left as a man speaks to camera on the right. A purple box slides from the timing column to the cost column, a lightning icon morphs into a dollar sign, then the highlighted row jumps to the top and the rest dim.

On-screen text: “PHONEBENCH · RESULTS”, “PhoneBench Alpha 1”

Techniques: Split screen, Morph

Camera: Static

07 · Product intro · 0:30–0:40

A black results card on the left shows PhoneBench Alpha 1 figures, 72.3% score, 331ms TTFAT and $0.0025 per minute, beside a man in a black t-shirt talking to camera on the right. The numbers fade in at the start; the presenter panel widens at the end.

On-screen text: “PHONEBENCH · RESULTS”, “PhoneBench Alpha 1”, “Pipecat PhoneLLM 30B Alpha 1”, “SCORE (T = 0.0)”, “72.3%”, “wins + draws vs K=16 tournament reference”, “TTFAT · P50”, “331ms”, “p95 ~600ms · floor 197ms”, “COST / MIN', '$0.0025', 'dedicated on Modal (H200)'”

Techniques: Split screen

Camera: Pan

08 · Problem · 0:40–0:42

A man in a black t-shirt faces the camera, raising both open hands in a stop gesture, then lowers them as he keeps talking. A blurred monitor behind him shows lines of code and a glowing blue circular logo.

On-screen text: “PI”

Camera: Pull out

09 · Problem · 0:42–0:44

A man in a black t-shirt speaks to the camera in a dim room, framed in a static medium close-up. Behind him a blurred blue circular graphic glows on the left and a monitor of green code lines sits on the right.

Camera: Static

10 · Problem · 0:44–0:47

A black line-drawn cat M logo draws itself in and scales up to the center of a white field carrying a faint rippling halftone dot pattern.

Camera: Static

11 · Problem · 0:47–1:04

A man in a black t-shirt speaks to camera in a locked-off medium close-up, centered before a large blurred monitor showing code and a blue ring logo. He gestures lightly while talking. At the end the frame slides left, leaving a black bar on the right.

Techniques: Masking

Camera: Pan

12 · Problem · 1:04–1:21

A man in a black tee speaks at camera left while a dark session panel right types a task, tool calls and a patch summary, its token counter climbing. The panel exits, leaving him full frame, then a white pill with cycling status text slides up over the dimmed shot.

On-screen text: “Mulling it over...”, “Weighing options...”, “Considering angles...”, “Double-checking...”

Techniques: Split screen, Kinetic type

Camera: Push in

13 · Problem · 1:21–1:35

A man in a black t-shirt sits centered, speaking to camera and gesturing with both hands. Behind him a left monitor glows with a blue ring logo and a right monitor shows blurred lines of code, both softly out of focus. The framing stays locked and steady throughout.

Camera: Static

14 · Problem · 1:35–1:42

Split screen: a man talks to camera on the left in a dark, code-lit room; on the right a vertical line joins a user icon to a robot icon. A latency figure counts from 0 to 2,016 ms as a dot drops the line and the robot spins a ring, labeled reasoning model.

On-screen text: “latency”, “187 ms”, “142 ms”, “236 ms”, “0 ms”, “616 ms”, “thinking...”, “reasoning model”, “2,016 ms”

Techniques: Split screen

Camera: Static

15 · Problem · 1:42–1:43

On black, a vertical line links two person icons to a robot icon below, with sparkles beside each node. A latency counter ticks up from 3,316 to 3,549 ms beside the bottom node, which carries the label reasoning model and a thinking... caption, then the frame cuts to black.

On-screen text: “latency”, “3,316 ms”, “reasoning model”, “thinking...”

Camera: Static

16 · Problem · 1:43–1:44

A man in a black t-shirt speaks to camera, framed chest-up, in a room with two monitors behind him, one showing a blue circular graphic and one showing code. The shot fades in from black.

Camera: Pan

17 · Problem · 1:44–1:55

Split screen: a dark benchmark table on the left shows model rows with score bars; its purple highlight slides from PhoneLLM 30B Alpha 1 to GPT-4.1. Right, a man in a black tee talks to camera before a blurred server rack. The frame dissolves in from a full close-up.

Techniques: Split screen

Camera: Static

18 · Problem · 1:55–1:56

Split screen: a man in a black shirt talks on the right in a dark room lit by green code; on the left a benchmark table of model names and scores clears out and a new two-column header row fades in.

Techniques: Split screen

Camera: Static

19 · Problem · 1:56–2:04

Split screen: a black table of model names and token counts builds row by row on the left, numbers ticking up, then dims out, while a man in a black tee talks to camera on the right against a blurred code wall.

Techniques: Split screen

Camera: Static

20 · Problem · 2:04–2:05

A man in a black t-shirt talks to the camera, centered in frame, with two out-of-focus monitors behind him, one showing a glowing blue ring and the other lines of code. He gestures with his hands as he speaks. The camera holds a static medium close-up.

Camera: Pan

21 · Feature reveal · 2:05–2:11

Split screen: a man talks on the left before a dark code wall. On the right, a cursor clicks the Reasoning dropdown, a menu of Low, Medium, High, Max and an Off toggle opens, the cursor flips the toggle, and the panel tilts in 3D, fading to the off toggle.

Techniques: Split screen, Cursor demo, 3D mockup

Camera: Static

22 · Feature reveal · 2:11–2:11

A split screen shows a man in a black GitHub tee on the left against a dark code backdrop, while a toggle switch flips on the right. The knob slides right as radiating lines collapse into concentric rings.

Techniques: Split screen, Morph

Camera: Static

23 · Feature reveal · 2:11–2:12

A single toggle switch with a light gray knob on a dark rounded track sits right of center on a black background. The icon holds still, then dims and fades out to black over the short shot.

Camera: Static

24 · Feature reveal · 2:12–2:21

A man in a black t-shirt stands centered and speaks to camera, framed from the chest up. Two large monitors fill the background, one showing a blue circular mark, the other lines of code. He raises both hands in a shrug near the end. The shot stays locked off.

On-screen text: “1”

Camera: Static

25 · Feature reveal · 2:21–2:24

A man in a black t-shirt speaks to the camera, centered in a medium close-up. A blurred blue circular logo with white squares sits left, a code editor glows right. He raises a hand to gesture mid-sentence.

Camera: Static

26 · Feature reveal · 2:24–2:25

A man in a black shirt sits centered before two dark monitors, one with a blue ring, one with code. The frame splits: a left black panel holds two numbered circles and the word EVALUATE, while a vertical line and scan lines cross his face on the right.

On-screen text: “1”, “2”, “EVALUATE”

Techniques: Split screen

Camera: Pan

27 · Feature reveal · 2:25–2:50

Split screen: a man in a black tee speaks on the right against blurred terminal screens. On the left, a six-node pipeline labeled EVALUATE to OBSERVE lights up in sequence with a purple glow, then fades into a bar chart that grows upward above two buttons, EVALS and BENCHMARKS.

On-screen text: “EVALUATE”, “DATA”, “VERIFY”, “TRAIN”, “DEPLOY”, “OBSERVE”, “EVALS”, “BENCHMARKS”

Techniques: Split screen, Kinetic type

Camera: Static

28 · Feature reveal · 2:50–2:58

Split screen: a man in a black tee talks on the right against a code backdrop, while on the left a purple bar chart gives way to a glowing sparkle tile that multiplies into a ring of six tiles, then fades to black.

On-screen text: “1”, “2”, “×”

Techniques: Split screen

Camera: Static

29 · CTA · 2:58–3:24

A man in a black t-shirt speaks to camera before a blue orb and code screen. At 4s he shrinks into a right panel over a dark dot-wave field, and two white URL cards labeled DOWNLOAD MODEL and CODE slide in on the left.

On-screen text: “DOWNLOAD MODEL”, “huggingface.co/pipecat-ai/phonellm-alpha-1”, “CODE”, “github.com/pipecat-ai”

Techniques: Split screen, Kinetic type

Camera: Static

30 · Outro · 3:24–3:25

A man in a dark t-shirt holds a neutral pose on the right over a blue abstract backdrop. A dark halftone dot field expands from the left across the frame, masking him out, while a small line logo drifts to center.

Techniques: Masking

Camera: Static

31 · Outro · 3:25–3:28

A white line-drawn cat logo, two pointed ears and dot eyes, holds center on a dark background of fine dots that ripple in slow waves like cloth, the mark brightening from gray to white over the first second.

Camera: Static