21 April 2026

Intelligence Pipeline — Inaugural Report

Signals detected, model landscape shifts, and actionable infrastructure intelligence for the week of 15–21 April 2026.

Executive Summary

The AI model landscape is experiencing a compression cycle: tier-2 providers are merging capabilities while frontier models push context windows past 1M tokens. Locally, our Ollama Cloud tier routing executed 847 queries this week with a 99.2% completion rate. The NecroSwarm v1.5.2 council architecture is stable across all 10 dimensions. Two signals warrant immediate attention: devstral-2:123b has emerged as a reliable backup for complex coding tasks, and deepseek-v3.2 appeared on the Ollama Cloud roster with no official changelog — potential unannounced release.

Signals

Detected Intelligence Signals

Automated signal extraction from model telemetry, GitHub activity, and news feeds.

Model Release

deepseek-v3.2: Unannounced on Ollama Cloud

New model ID appeared in Ollama Cloud API with no release notes or changelog. Performance benchmarks unclear. May be a hotfix release or incremental update to v3.1:671b.

Critical Confidence: 78%

Infrastructure

devstral-2:123b Promoted to Backup T2

devstral-2:123b:cloud has been validated as the official backup for T2 complex routing. Previously untested at this tier. Early signals show strong code-generation with 91% pass@2 on HumanEval.

Actioned Confidence: 94%

Trend

Ollama Cloud T1 Expansion: 5→6 Models Expected

Rumored addition of a new T1 model to the :cloud roster. Ollama Inc. has not confirmed, but API probe returns a masked placeholder at the T1 slot. Likely a coding-specialized model.

Monitor Confidence: 62%

Security

Stale OLLAMA_API_KEY in .env

The .env file still contains the old OLLAMA_API_KEY which was superseded by OLLAMA_CLOUD_API_KEY. No current exploit path, but latent confusion vector. Also: _resolve_task_provider_model() ignores api_key_env from config — auth bypass risk.

Fix Needed Confidence: 100%

Model Landscape

Ollama Cloud Tier Routing — This Week

Tier	Models	Queries	Success	Avg Latency
T1 — Frontier	kimi-k2.5, deepseek-v3.1:671b, glm-5.1, qwen3-coder:480b, mistral-large-3:675b	423	99.5%	3.2s
T2 — Professional	minimax-m2.5, gemma4:31b, devstral-2:123b	201	98.5%	1.8s
T3 — Utility	gemma3, ministral-3, nemotron-nano	178	99.4%	0.9s
Think — Chain	kimi-k2:1t	45	97.8%	8.4s

Infrastructure

Active Projects & Deployments

Project	Platform	Status	Last Deploy
radikalekonzept.org	CF Pages (rk-landing)	Live	21 Apr 2026
intelligence.teoh.my	CF Pages (intelligence-teoh)	Live	21 Apr 2026
VoidTether	GitHub (0x-wzw/voidtether)	Alpha	—
NecroSwarm v1.5.2	Local skill	Stable	Ongoing

Actions

Recommended Actions This Cycle

1 Remove stale OLLAMA_API_KEY from .env and verify OLLAMA_CLOUD_API_KEY is the single source of truth. Patch _resolve_task_provider_model() to respect api_key_env from config.

2 Benchmark deepseek-v3.2 against deepseek-v3.1:671b on standard eval suite. If improvements confirmed, consider promoting to T1.

3 Monitor T1 expansion slot — set up automated API probe to detect when the masked model is formally listed. Trigger council analysis when it appears.

4 Harden Cloudflare Pages deploys — current pipeline is manual wrangler. Consider adding a CI/CD guardrail via GitHub Actions for automated deploys on push to main.