FI daily research brief 2026-11-03 (activity-core rhythm)

briefs/2026/11/2026-11-03.md

@@ -0,0 +1,51 @@
+---
+date: 2026-11-03
+timezone: Europe/Berlin
+author: rein-aharness
+status: final
+sources_checked:
+  - docs/sources-allowlist.md
+  - automated field scan via llm-connect
+---
+
+# Freedom Intelligence Daily Brief — 2026-11-03
+
+> Automated brief (activity-core schedule → rein-aharness). Deltas only.
+
+## Headline deltas
+
+- DeepSeek-AI released `DeepSeek-V4-Flash-0731-Instruct`, an instruction-tuned variant of the existing S1 strategic model.
+- The new Instruct model is ~304B / ~13B-active MIT MoE, consistent with the base model's architecture.
+- This release provides an official instruction-tuned version for the already collected `DeepSeek-V4-Flash-0731`.
+- No new frontier API releases, open-weight models from other orgs, training methods, or harness updates were observed.
+
+## Frontier & commercial (axis A)
+
+*(none)*
+
+## Edge / local / open (axis B)
+
+| Item | Delta | Sources | Lab relevance |
+| ---- | ----- | ------- | ------------- |
+| DeepSeek-V4-Flash-0731-Instruct released | New instruction-tuned variant of the DeepSeek-V4-Flash-0731 base model. This is a ~304B / ~13B-active MIT MoE (~167 GiB) model, not a 12B dense model. It provides an official instruction-following cap | https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731-Instruct | Provides an officially instruction-tuned version of our S1 strategic model, potentially improving out-of-the-box performance for chat and agentic tasks without additional fine-tuning. |
+
+## Training & specialization (axis C)
+
+*(none)*
+
+## Harness & fleet (axis D)
+
+*(none)*
+
+## Collection candidates
+
+| id | org | name | priority | reason | approx size | license |
+| -- | --- | ---- | -------- | ------ | ----------- | ------- |
+| deepseek-ai__DeepSeek-V4-Flash-0731-Instruct | deepseek-ai | DeepSeek-V4-Flash-0731-Instruct | high | Instruction-tuned variant of an already collected S1 strategic model (DeepSeek-V4-Flash-0731). Offers direct utility for chat and agentic workflows. License is MIT, consistent with the base model. | ~167 GiB (MoE, same as base) | MIT |
+
+## Lab implications
+
+- Prioritize collection of `DeepSeek-V4-Flash-0731-Instruct` to leverage the instruction-tuned capabilities of our S1 strategic model. This should be a high priority given the existing base model collection.
+- Re-evaluate the role of `Qwen/Qwen3.5-7B-Instruct` as a primary actionable item, as `DeepSeek-V4-Flash-0731-Instruct` offers a significantly more capable, albeit larger, MIT-licensed instruct model.
+- Update `inventory/catalog/deepseek-ai__DeepSeek-V4-Flash-0731__strategic.yaml` to reflect the availability of an instruction-tuned sibling and consider linking the two entries.
+- No immediate changes to `llm-connect` routing or other strategic reserve planning are required beyond the new collection.