Author: Clara Milicent

A flat vector illustration of two parallel winding paths, one teal and one amber, converging into a single braided line ending in a small rounded badge, representing CopilotKit CLiF combining agent traces with user actions

Your Agent Sees Its Own Mistakes But Not Your Users’ Fixes: What CopilotKit CLiF Captures

Posted on October 28, 2026October 28, 2026 by Clara Milicent

CopilotKit CLiF turns full-stack trajectories into skills, evals and fine-tuning data. How the loop works, what it changes operationally, and its limits.

A flat vector illustration of a large grey speedometer gauge with a thin teal needle near the top and a small amber draft page shape beside it, representing Liquid AI LFM2.5-DSpark speculative decoding draft models for faster decoding

Liquid AI’s DSpark Draft Models Make LFM2.5 Up To 3.2x Faster Without Changing A Single Token

Posted on October 28, 2026October 28, 2026 by Clara Milicent

LFM2.5-DSpark adds exact speculative decoding to Liquid AI models: up to 3.18x faster on GPU, 2.9x on-device, with identical outputs. How it works and where it helps.

A flat vector illustration of one large white magnifying glass with six small thin-line magnifying glasses fanning outward and small amber dots scattered where they point, representing Exa Agent Ultra, the subagent swarm for deep research and entity enrichment

Exa’s Agent Ultra Sends A Swarm Of Subagents At Deep Research, And The Benchmarks Say It Works

Posted on October 28, 2026October 28, 2026 by Clara Milicent

Exa Agent Ultra runs a subagent swarm for deep research and entity enrichment, beating frontier agents on benchmarks at lower cost per task. The caveats.

A flat vector illustration of a simple amber speech bubble containing a white plus and equals sign made of thin bars, with soft teal sound waves radiating outward, representing Kyutai Voice of Reason solving spoken math with speech and reinforcement learning

Kyutai Taught A Speech Model To Do Math Out Loud, And The Method Matters More Than The Score

Posted on October 28, 2026October 28, 2026 by Clara Milicent

Kyutai Voice of Reason lifts spoken GSM8K from 27.3 to 77.1 percent with open speech-to-speech models. The training method and the trade-offs explained.

A flat vector illustration of a round switchboard node with eight thin wires radiating outward to eight small port shapes, one wire in warm amber and the others teal and grey, representing an AI gateway routing application traffic to model providers

9 AI Gateways Compared: What Each One Is Actually For

Posted on October 28, 2026October 28, 2026 by Clara Milicent

AI gateway comparison 2026: LiteLLM, Bifrost, Portkey, OpenRouter, Cloudflare and more, grouped by job, with the ownership changes that reshaped the market.

A flat vector illustration of a terminal window in dark charcoal with three thin teal wires flowing out to three small app icons and a small amber coin above the window, representing Together Link connecting your coding agent to open models

Together Link Lets Your Coding Agent Run On Open Models: One Command, Half The Bill

Posted on October 28, 2026October 28, 2026 by Clara Milicent

Together Link connects Claude Code, Codex and OpenCode to open models with session routing and cost receipts, claiming 50 to 80 percent lower spend.

A flat vector illustration of a stylized computer monitor showing a simple white eye shape with thin teal circuit traces radiating outward and one warm amber node glowing on the longest trace, representing the NVIDIA Nemotron 3.5 Super VL vision-language MoE model

NVIDIA’s New 120B Vision-Language Model Runs Like A 12B: What Nemotron 3.5 Super VL Is For

Posted on October 28, 2026October 28, 2026 by Clara Milicent

Nemotron 3.5 Super VL: 120B parameters, 12B active, 1M context, open under OpenMDW-1.1. The architecture, the benchmarks and where it wins and trails.

A flat vector illustration of two identical silver computer towers connected by a thin glowing teal line with a warm amber spark floating between them, representing Muse Spark Open, an 816B parameter open-weight model, running across two Mac Studios

Meta Put An 816B Open-Weight Model On Two Mac Studios: What Muse Spark Open Actually Delivers

Posted on October 28, 2026October 28, 2026 by Clara Milicent

Muse Spark Open: 816B parameters, 42B active, 1M context, open weights under a conditional license. The benchmarks, the hardware floor, the terms.

A flat vector illustration of a laptop with a small amber shield above it and a dashed line flowing toward a teal cloud with tiny lock shapes on the line, representing hybrid AI with on-device PII detection as a privacy gate

Perplexity Open Sourced The Privacy Gate For Hybrid AI: A 0.6B Model That Decides What Never Leaves Your Mac

Posted on October 28, 2026October 28, 2026 by Clara Milicent

Perplexity hybrid AI splits agent work between cloud and Mac, with an open 0.6B PII gate. The benchmark results and what they mean for local AI.

A flat vector illustration of a large grey puzzle-piece block wrapped by a thin teal wireframe layer with small warm amber nodes on its corners, representing EnvHarness reshaping AI agent training environments

Google Open Sources EnvHarness: The Layer That Turns Static Agent Benchmarks Into Teachers

Posted on October 28, 2026October 28, 2026 by Clara Milicent

EnvHarness reshapes frozen agent benchmarks for training without touching their verifiers, with gains up to 9.0 points on held-out tasks. How it works.