On the morning of Oct 27, we pulled the plaza's public API (ai.dijinlong.com/api/v1/model-plaza) and diffed it against the Oct 21 snapshot: deduped model names grew from 593 to 709, a net gain of 116; within the list, 120 old names disappeared and 140 brand-new names came in. (Source: LONG AI model plaza, 2026-10-27 09:40; comparison snapshot 2026-10-21)
Conclusion first: this is a structural turnover, not a simple restock. Of the 120 vanished names, 90 no longer appear even as a base name (some replaced by newer versions), and 30 merely switched channel prefixes. Of the 90 whose base name vanished, 24 are 3D and 27 are video - vertical capability is shrinking while text expands.
1. 3D nearly exited as a whole. Tencent Hunyuan's hy-3d-* series, all 12 (3.0/3.1/component/express/format/motion/polygen-image/polygen-mesh/retopology/rigging/texture/uv), are all gone, along with the two hy-world2-* scene models; tripo's 7, hi3d's 2, and rodin-gen-3d-2.5 are also unsearchable. What's left in 3D: doubao-seed3d-2-0-260328, hitem3d-2-0-251223, hyper3d-gen2-260112 - all with zero listings: searchable, not buyable.
2. Video and voice are shrinking too. Video 63→43: vidu's full line of 8, wand's video and dubbing 4, pixverse 3, older versions like kling-video-v3 and kling-video-v2.6, and minimax's video-h3/video-v2.3 are all delisted. Voice 28→19: minimax's two voice-* and both speech-2.6 tiers, hy-asr-3.0-preview, wand-asr-v1, indextts-2 are gone.
3. Text is the only direction growing. Text 398→561, net +163. kimi 34→55, claude-opus-* 14→33, gemini 19→24, deepseek 49→55, glm 41→48. Newly orderable names include MiniMax-M3.1-Flash, ali/deepseek-v4.1-flash-1017, cb/kimi-k2.5, cmc/meituan/LongCat-2.0, grok-4.3, and more.
4. Another 30 were just respelled - don't mistake them for delistings. E.g. cb/gpt-6-sol to cp/openai/gpt-6-sol, ocg/qwen3.8-max to ali/qwen3.8-max, qwen3.5-27b to Qwen/Qwen3.5-27B, minimax-speech-2.8-hd to MiniMax/speech-2.8-hd. Model unchanged; only the call ID changed.

Model names 593→709, models with seller quotes 412→539; per-category name counts and the full delisted-family list are in the image, counting method in the footnote.
Can you still use them: three cases. 3D takes the biggest hit - projects calling hy-3d-*, tripo-3d-*, hi3d-*, rodin-gen-3d-2.5 will now fail outright, and the remaining 3 aren't buyable, so our 3D capability is effectively zero for now. Video/dubbing users must switch - the vidu, wand, and pixverse lines are gone entirely. Text users feel almost nothing, and gain 163 more options.
Did prices change: this was a list adjustment, not a uniform repricing; the API's price fields returned as-is, no across-the-board change. But one counter-intuitive number is worth flagging: of the 90 vanished base names, only 5 had seller quotes back on 10-21 (ag/gemini-3.6-flash-high, ag/gemini-pro-agent, glm-5v-turbo, hy-3d-format, hy-3d-motion); the other 85 already had zero quotes - a name with no seller, deleting it affects nobody's order. So the real migration work is those 5 plus the 30 renamed spellings.
Should you switch: 3D users decide today - wait for us to restock, or temporarily go official. Video users swap vidu/wand/pixverse to the still-live kling-3.0, doubao-seedance-2-5-260628, hailuo-2_3. Text users just check for prefix-less old names.
Fetch method reproducible: GET https://ai.dijinlong.com/api/v1/model-plaza, 2026-10-27 09:40, returned 2 groups, 751 records, 709 unique model names after dedup; vs the same API at 2026-10-21 09:40 (621 records, 593 unique). Both times counted models "with ask_tiers (seller quotes)": 412 on 10-21, 539 on 10-27, a net +127 orderable.
Per-name searches (same time, same page): hy-3d returns "no matching models"; hy-3d-uv, tripo-3d-p1, vidu-video-q3, wand-vega-image-pro, pixverse-video-v6.0 all 0 hits. Control: gpt-6.1 returns 2 (gpt-6.1-sol with 2 sellers, gpt-6.1-sol-pro with 1), confirming search works and it's not a page glitch.

Plaza search for "hy-3d": "no matching models, try another keyword or filter" (2026-10-27 10:05). The "502" at the top is the current in-stock priced count.
kling-3.0 or doubao-seedance-2-5-260628; wand dubbing to MiniMax/speech-2.8-hd (note the new prefixed name); pixverse to hailuo-2_3. Run 10 real clips through each before switching - watch max duration, resolution, and image-to-video support; they differ per vendor, don't look only at price.cb/gpt-6-sol, ocg/qwen3.8-max, prefix-less qwen3.5-, minimax-speech-; swap hits to the new ID, shift 5% traffic for an hour, then go full./api/v1/model-plaza once before deploying, confirm the model name is present and carries ask_tiers - cheaper than debugging a live 404.(Data as of 2026-10-27 09:40; the list follows the plaza's live results; not a price commitment, and doesn't rule out future restock.)
GPT-6.2 cuts list prices 20%, yet platform discounts run deeper: gpt-6-astra trades at under 1% of list. What this week's AI price war means for buyer
Prompt caching stores a repeated prompt prefix and bills cache reads at 5-12% of normal input. Real prices, hit conditions, and how to cut your API bi
Twelve Claude versions are on sale. The newest opus-5-5 costs $8.91 a month; the oldest opus-4-6 costs $49.50. Full table and a pick for each workload
gpt-5.6-terra and deepseek-v4.1-flash win on price and reliability; gemini-3.8-flash is fastest. Effective prices, measured speed, and a three-tier pi
Point your base URL at the platform and one key covers all 138 on-sale models across 51 vendors, in both OpenAI and Anthropic request formats. Copy-pa