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.

AI-for-materials company Deep Principle released and open-sourced its first 753B general science model Hydrogen-1.0 on Nov 2, weights under MIT on Hugging Face and ModelScope. Key points:
Take: this company put "open base + in-house full-pipeline data" front and center, letting university labs and small teams truly self-host. But be precise - all scores are self-reported, and some comparisons excluded GPT-6-Astra and Opus 5.5 for "too many refusals," making the control set selective; "self-claimed above Astra" is not third-party sign-off.
| Model / Move | Date | Type | Highlights |
|---|---|---|---|
| Qwen4 repo leak (Alibaba) | 11-3 | Leak + tease (not a release) | qwen4-max, qwen4-coder-plus, and qwen4-omni-120b appeared in the Qwen Code repo, hinting the new flagship is near; management the same day debunked "API-only" rumors, reaffirming 4.0 and 5.0 stay open. No official model name yet - don't trust any score rumor. |
| MiniMax M3.1 Flash community quantization | 11-3 | Open-source ecosystem | Ten days after the weights opened, community developers compressed the 428B model from 252.4GB to 93GB (quantization-aware training + sparse attention), with MiniMax publicly thanking them - a "smaller and smaller" positive loop for open weights. |
| Sawara (Sakana AI) | 11-3 | Teaser | Japan's Sakana AI announces its first small-model plan "Sawara," public only in spring 2027 - a distant teaser. |
Today's keyword is "openness going granular." Hydrogen-1.0 open-sources the whole lab process including failures, Google opens its Android tools to rivals' agents, Qwen4 gets "spoiled" by its own repo before launch - openness is no longer a stage slogan but a contest of who spreads their cards thinner and delivers faster.
(Data from official announcements and public reports, 2026-11-04; 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