跪拜 Guibai
← All articles
AI Programming · OpenAI · Artificial Intelligence

The August 2026 AI Build List: What to Actually Use for Coding, Images, and Video

By stormzhangV ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

The AI tooling landscape has shifted enough in one month that a subscription stack from three months ago is already suboptimal. Developers betting on a single vendor or ignoring domestic Chinese models risk paying more for less capable tools, especially in video where open-source H3 undercuts proprietary pricing by two-thirds.

Summary

Claude Code's once-commanding lead has evaporated; Anthropic's slow progress and weak ecosystem make it skippable for new users. OpenAI's Codex paired with GPT-5.6 Sol now dominates programming, backed by an ecosystem where GPT Image 2 still holds the text-to-image crown and voice remains unmatched. For developers inside China who can't access Codex, Kimi Code with the K3 model is the surprise contender, punching far above its perceived weight in complex tasks despite compute shortages.

In video, ByteDance's Seedance 2.5 delivers 30-second native clips with timestamp-level editing, pushing AI video from raw素材 to finished output, though at a steep price. MiniMax H3 counters with fully open-source weights, 2K 15-second generation with native stereo sound, and API pricing at one-third of mainstream models. The list closes with persona-based picks: Codex for professional devs, Workbuddy for office workers, GPT Image 2 for designers, and a practical note that combining Motion Graphics engineering with video models slashes production costs far below pure generation.

Takeaways
Claude Code is no longer a must-have; Anthropic's Opus 4.7 through Opus 5 have been poorly received, and the company's ecosystem gap is widening.
Codex plus GPT-5.6 Sol is the strongest programming combination, with frequent subscription resets making it the best value play right now.
GPT Image 2 has held the number-one text-to-image arena spot for nearly four months since its April 21 release.
Grok's Imagine Image 2.0 ranks second globally on both text-to-image and image editing leaderboards, with output speed described as machine-gun fast.
Kimi Code with the K3 model handles complex programming tasks competitively against Codex, though compute shortages make subscriptions hard to get and generation slower.
Seedance 2.5 generates 30 seconds of native video in one pass with timestamp-precise editing, while the industry average remains 5 to 15 seconds.
MiniMax H3 is fully open-source, outputs 2K 15-second video with native stereo sound, and costs $0.13 per second at 2K resolution via API.
Combining Motion Graphics engineering with video models can drastically cut production costs compared to relying solely on generative video models.
Conclusions

Anthropic's trajectory from cliff-like leader to skippable option in three months suggests that early-mover advantage in AI tooling decays fast when ecosystem breadth matters more than raw model quality.

OpenAI's frequent Codex subscription resets are a deliberate user-acquisition tactic; the real test is whether capability stays ahead once the resets slow down as the user base grows.

Kimi K3 being underestimated by its domestic audience while matching Codex in complex tasks points to a persistent credibility gap for Chinese models, even when performance warrants serious consideration.

Grok's speed advantage and Musk's capital and compute resources position it as a dark horse across programming, image, and video, but it remains unproven as a primary tool.

MiniMax open-sourcing H3 weights immediately puts downward pressure on the entire video generation market; the marginal cost of local inference is electricity, which reshapes the economics for teams with GPUs.

Concepts & terms
Motion Graphics vs. Creative Asset Video
Motion Graphics are videos produced through code-driven engineering, giving frame-level control without generative AI. Creative asset videos rely on large video models to generate footage, which is expensive and requires multiple attempts (抽卡, 'gacha pulls') to get usable results. Combining both approaches can maintain quality while slashing costs.
Codex subscription resets
OpenAI periodically resets usage quotas or offers fresh subscription terms for Codex, effectively giving users renewed access to compute-heavy features. This is a promotional tactic to attract and retain users, but the frequency is expected to decrease as the user base stabilizes.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗