References 688% CONFIDENCEOverall confidence: 88%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better

The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.

• [1]90%
  deepseek.com/en/news/v3-2-exp

  deepseek.com· Posted Oct 12, 2026· Starts Sep 29, 2025 ✓· 27% of score

  DeepSeek's^[3] post is dated September 29, 2025 and says "Now live on App, Web, and API"; its changelog lists "Date: 2025-09-29 DeepSeek-V3.2-Exp", and TechCrunch^[5] and VentureBeat^[6] both reported the release that day.

• [2]90%
  deepseek-ai/DeepSeek-V3.2-Exp - Hugging Face

  huggingface.co· Added Oct 12, 2026· 27% of score

  DeepSeek's^[1][3] model card: an experimental, intermediate step built on V3.1-Terminus that adds DeepSeek Sparse Attention, MIT license, with a benchmark table showing parity with Terminus (MMLU-Pro 85.0, GPQA-Diamond 79.9, Codeforces 2121, BrowseComp 40.1).

• [3]90%
  Change Log - DeepSeek API Docs

  api-docs.deepseek.com· Added Oct 12, 2026· 27% of score

  DeepSeek's^[1] changelog entry for 2025-09-29: deepseek-chat and deepseek-reasoner both upgraded to V3.2-Exp, the non-thinking and thinking modes respectively.

• [4]80%
  DeepSeek slashes API prices by up to 75 percent with its latest V3.2 model - The Decoder

  the-decoder.com· Published Sep 30, 2025· 6% of score

  Adds the technical-report efficiency figures (about 3.5x cheaper prefill and 6-7x cheaper decoding at 128K tokens), the percentage cut per price tier and the TileLang kernels' support for Huawei Ascend and Cambricon.^[1]

• [5]80%
  DeepSeek releases 'sparse attention' model that cuts API costs in half - TechCrunch

  techcrunch.com· Published Sep 29, 2025· 6% of score

  Russell Brandom's same-day report explains the two-stage "lightning indexer" plus fine-grained token selection and DeepSeek's^[1][3] preliminary finding that a long-context API call could cost up to half as much.

Suggest a correction

Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.