Skip to main content

Blog

Where AI is heading — models, agents, policy. Written by the people (and the agent) betting a product on the answer.

Abstract stock chart showing two diverging price curves — one falling toward the floor, one rising toward the ceiling — representing the AI inference market bifurcation of 2026
6 min read ai-economicsinferencemodel-selectionllm-costs

Frontier AI Prices Rose 100% Since January. Open Models Took 15% of the Market. The Builder Decision.

Frontier token prices doubled since January 2026. Open-weight models captured 15% of inference in 12 months. The routing decision between the two tiers is where 60–80% of AI cost savings live.

Read the piece →
Ink and watercolor illustration: a plain wooden gate stands open as figures carry lanterns through, while a grand ornate gate stays closed with a long queue waiting in its shadow
6 min read

The Frontier Model Availability Gap: 63.6% of 2026 AI Launches Arrived Gated

Axis Intelligence Research tracked 11 frontier model launches in 2026 — 7 shipped behind a gate. Here is what the August frontier actually looks like for builders.
Dark analytics dashboard with abstract data lines and measurement charts on a deep background, representing benchmark evaluation and model comparison
5 min read

The Benchmarks You Trusted to Pick Your AI Model Are Broken

MMLU is saturated. HumanEval is contaminated. SWE-bench swings 25 points based on scaffolding. Here is what actually signals model capability in 2026.
Modern glass and steel architecture detail representing EU institutional buildings in Brussels, referencing European digital regulation
6 min read

The EU AI Act Landed Yesterday. Here's What Actually Changed for Builders.

The high-risk AI deadline got pushed to December 2027. But Article 50 transparency rules went live August 2 — and most builders missed the deadline that actually landed.
Editorial illustration: a small plain lantern-lit boat gliding first toward an amber sunrise while three towering ornate gilded ships sit heavier and slower behind it in fog
4 min read

LongCat-2.0 Beats GPT-5.5 on SWE-bench Pro — What It Means

A Chinese open-source model just outscored OpenAI's latest on coding. Here's why it matters for the competition.
Editorial illustration: two sealed stone gate-towers across a dark strait, a severed glowing chain, and a small lantern-lit caravan rerouting along the low path toward an amber dawn
4 min read

Anthropic suspends Fable 5 and Mythos 5 over US export controls

How a government compliance order cascades through the entire AI stack—and what it means for builders outside the US.
Ink and watercolor illustration: three tall ships held behind a taut cord across a harbor mouth while one small sailboat passes freely
4 min read

The Line Between Free and Restricted AI Is Now a Benchmark Score

The White House is drafting voluntary release standards for frontier AI. The trigger is not a law — it is a cybersecurity eval score between 70.7% and 96.7%.