Shot list — deepseek-v4-preview

Generated 2026-04-28 14:48 PDT · 56 cues · alignment 100% · 25fps · timeline starts 01:00:00:00

01:00:00:00Talking head[TALKING HEAD — hook]
… One day after OpenAI released GPT-5.5, DeepSeek struck back. …
01:00:03:21CUT[CUT]
… On April 24, 2026, the Chinese AI lab announced DeepSeek V4 Preview, releasing t …
01:00:03:21VOICEOVER[VOICEOVER — scene 1]
… On April 24, 2026, the Chinese AI lab announced DeepSeek V4 Preview, releasing t …
01:00:03:21B-roll[B-ROLL: ai-abstract]
… On April 24, 2026, the Chinese AI lab announced DeepSeek V4 Preview, releasing t …
01:00:31:04Asset[B-ROLL: company-logo:DeepSeek]
… Twelve months ago, DeepSeek V3's surprise arrival shook the global AI industry a …
01:01:03:06STAT[STAT CARD: "V3 debut: April 2025 shockwave"]
… V4-Pro is a massive mixture-of-experts model with 1.6 trillion total parameters. …
01:01:03:06B-roll[B-ROLL: data-center]
… V4-Pro is a massive mixture-of-experts model with 1.6 trillion total parameters. …
01:01:38:14STAT[STAT CARD: "1600B total params in V4-Pro"]
… A 1 million token context window is not just a bigger number. It is a different …
01:01:38:14STAT[STAT CARD: "V4-Pro: 49B params active per pass"]
… A 1 million token context window is not just a bigger number. It is a different …
01:01:38:14STAT[STAT CARD: "V4-Flash: 284B params, 13B active"]
… A 1 million token context window is not just a bigger number. It is a different …
01:01:38:14B-roll[B-ROLL: code-terminal]
… A 1 million token context window is not just a bigger number. It is a different …
01:02:10:15B-roll[B-ROLL: data-center]
… DeepSeek published the model weights on Hugging Face. V4-Pro downloads at 865 gi …
01:02:40:24STAT[STAT CARD: "V4-Pro download: 865 GB model"]
… The benchmark numbers are where V4-Pro gets genuinely difficult to dismiss. Star …
01:02:40:24STAT[STAT CARD: "V4-Flash download: 160 GB model"]
… The benchmark numbers are where V4-Pro gets genuinely difficult to dismiss. Star …
01:02:40:24VOICEOVER END[/VOICEOVER]
… The benchmark numbers are where V4-Pro gets genuinely difficult to dismiss. Star …
01:02:40:24Talking head[TALKING HEAD — transition]
… The benchmark numbers are where V4-Pro gets genuinely difficult to dismiss. Star …
01:03:10:10STAT[STAT CARD: "SWE-bench: 80.6% vs 80.8% Claude"]
… On competitive programming, V4-Pro did not just approach the American frontier. …
01:03:10:10CUT[CUT]
… On competitive programming, V4-Pro did not just approach the American frontier. …
01:03:10:10VOICEOVER[VOICEOVER — scene 2]
… On competitive programming, V4-Pro did not just approach the American frontier. …
01:03:10:10B-roll[B-ROLL: finance-charts]
… On competitive programming, V4-Pro did not just approach the American frontier. …
01:03:45:04STAT[STAT CARD: "Codeforces: 3206 — #1 open source"]
… The science and mathematics results follow the same pattern. V4-Pro scored 90.1 …
01:03:45:04STAT[STAT CARD: "GPT-5.4 Codeforces rating: 3168"]
… The science and mathematics results follow the same pattern. V4-Pro scored 90.1 …
01:03:45:04Asset[B-ROLL: stills:science-publications]
… The science and mathematics results follow the same pattern. V4-Pro scored 90.1 …
01:04:13:04STAT[STAT CARD: "GPQA Diamond: 90.1% PhD-level"]
… Long-context performance holds up under structured testing. On MRCR 1M, a retrie …
01:04:13:04STAT[STAT CARD: "MMLU-Pro: 87.5% multidomain"]
… Long-context performance holds up under structured testing. On MRCR 1M, a retrie …
01:04:13:04STAT[STAT CARD: "HMMT Feb 2026 math competition"]
… Long-context performance holds up under structured testing. On MRCR 1M, a retrie …
01:04:13:04STAT[STAT CARD: "HMMT math: 95.2% — frontier tier"]
… Long-context performance holds up under structured testing. On MRCR 1M, a retrie …
01:04:13:04B-roll[B-ROLL: code-terminal]
… Long-context performance holds up under structured testing. On MRCR 1M, a retrie …
01:04:45:07STAT[STAT CARD: "MRCR 1M: 83.5% retrieval score"]
… On Terminal-Bench 2.0, a benchmark for autonomous coding agents that must naviga …
01:04:45:07Asset[B-ROLL: screen-capture:benchmark-chart]
… On Terminal-Bench 2.0, a benchmark for autonomous coding agents that must naviga …
01:05:32:11STAT[STAT CARD: "Terminal-Bench: 67.9% vs GPT 82.7%"]
… Against other open-source competitors, the story is less mixed. Alibaba's Qwen a …
01:05:32:11STAT[STAT CARD: "SWE-bench Pro: 55.4% harder tasks"]
… Against other open-source competitors, the story is less mixed. Alibaba's Qwen a …
01:05:32:11Asset[B-ROLL: company-logo:Alibaba]
… Against other open-source competitors, the story is less mixed. Alibaba's Qwen a …
01:06:03:15B-roll[B-ROLL: data-center]
… The architecture behind the performance is a genuine engineering achievement. De …
01:06:50:23STAT[STAT CARD: "27% of V3.2 FLOPs at 1M ctx"]
… The pricing makes that efficiency impossible to ignore. V4-Pro costs $1.74 per m …
01:06:50:23STAT[STAT CARD: "KV cache: only 10% of V3.2 needs"]
… The pricing makes that efficiency impossible to ignore. V4-Pro costs $1.74 per m …
01:06:50:23STAT[STAT CARD: "Trained on 32T tokens of data"]
… The pricing makes that efficiency impossible to ignore. V4-Pro costs $1.74 per m …
01:06:50:23B-roll[B-ROLL: finance-charts]
… The pricing makes that efficiency impossible to ignore. V4-Pro costs $1.74 per m …
01:07:42:24STAT[STAT CARD: "V4-Pro: $1.74/M in vs $5 GPT-5"]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24STAT[STAT CARD: "V4-Pro output: $3.48/M tokens"]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24STAT[STAT CARD: "GPT-5.5 output: $30/M per token"]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24STAT[STAT CARD: "Claude Opus output: $25/M tokens"]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24STAT[STAT CARD: "V4-Flash: $0.14/M input tokens"]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24STAT[STAT CARD: "V4-Flash output: $0.28/M tokens"]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24VOICEOVER END[/VOICEOVER]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:07:42:24Talking head[TALKING HEAD — transition]
… Markets responded immediately. On V4 launch day, shares of SMIC, the Chinese chi …
01:08:09:22CUT[CUT]
… V4 is DeepSeek's first release with explicit optimization for Huawei Ascend proc …
01:08:09:22VOICEOVER[VOICEOVER — scene 3]
… V4 is DeepSeek's first release with explicit optimization for Huawei Ascend proc …
01:08:09:22B-roll[B-ROLL: military]
… V4 is DeepSeek's first release with explicit optimization for Huawei Ascend proc …
01:08:35:06B-roll[B-ROLL: news-studio]
… Nvidia CEO Jensen Huang has said plainly what is at stake. His words: "The day t …
01:09:09:04B-roll[B-ROLL: courtroom]
… The IP dispute surrounding DeepSeek has intensified alongside V4. White House Of …
01:09:54:18STAT[STAT CARD: "24000 fake accounts traced"]
… DeepSeek has not provided a timeline for moving V4 out of Preview status. The AP …
01:09:54:18STAT[STAT CARD: "16M stolen Claude API queries"]
… DeepSeek has not provided a timeline for moving V4 out of Preview status. The AP …
01:09:54:18B-roll[B-ROLL: code-terminal]
… DeepSeek has not provided a timeline for moving V4 out of Preview status. The AP …
01:10:13:03VOICEOVER END[/VOICEOVER]
… The AI race we are living through has stopped looking like a pure technology com …
01:10:13:03Talking head[TALKING HEAD — sign-off]
… The AI race we are living through has stopped looking like a pure technology com …