This column aggregates the latest large-model releases and open-source drops from the past one to two days, cross-checked against official announcements and multiple media outlets. Performance figures distinguish between vendor claims and independently verified third-party results; anything not yet replicated is clearly marked. Illustrations are original.

The "Space Bunny" revealed: the anonymous Space Bunny that had been topping leaderboards and handling over 200 trillion tokens is officially confirmed as an early build of M3.1 Flash - run in live traffic for weeks before being given its formal name.

Availability: live on the API, M Plan, and MiniMax Code; weights open later - not open source yet. A Pro version has been teased.
| Model | Date | Highlights |
|---|---|---|
| MemT 1.0 (open source) | 10-20 | A Chinese team's "memory-native" model: persistent, updatable internal memory baked into computation. Flash 288B / Nano 36B, downloadable on Hugging Face. First or second across 11 memory benchmarks for parameterized-memory methods; Nano's long-history inference is 1.86x (128K) to 3.99x (256K) faster than full-context. |
| NVIDIA Nemotron 3.5 Super VL (open source) | 10-19/20 | 120B open weights for always-on enterprise agents (document intelligence, finance, insurance claims), trialed by EY, Salesforce, Perplexity and others. |
| NVIDIA Bolt x2 (open source) | 10-19/20 | On-device models running locally on RTX Spark AI PCs and DGX Spark workstations. |
| Mistral Timestral | 10-20 | A time-series foundation model in four sizes with multivariate probabilistic forecasting (not a language model; worth watching for forecasting teams). |
| China Telecom Xing4.0-29B-Intent (open source) | 10-20 | An intent-understanding model with open weights and 2.75M training samples, already deployed in telecom support and smart speakers. |
| StartLux-Decision (open source upgrade) | 10-20 | A dual-track decision model: Instant (0.8B-27B) for fast judgment plus AutoThink (4B-27B) for on-demand deep thought; claims to reset the record for decision models. |
| Daxiao ACE-Brain-0.6 | 10-19 | An embodied-AI base model fine-tuned from Qwen3.5-9B; vendor claims six spatial-perception benchmarks beat GPT-6 Astra. |


"Cheap, fast, long-running" Flash/agent models are the new battleground (MiniMax M3.1 Flash picks up the torch from DeepSeek V4.1-Flash), while "memory" has become a fresh lane (MemT 1.0 and Hy Memory 2.0 landing the same week).

(Data from official announcements and public reports, 2026-10-20; vendor claims and third-party results are marked separately, with unverified items noted in the text. Model capability affects selection, but actual pricing follows the model plaza in real time.)
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