Early build · scores computed by the deterministic engine from real, sourced developments (30 days) · weights v1.0-prior
Updated 06:00 UTC
next refresh in –
☰
Human moment
Vikshy Grade
Frame 214
An iPhone in Shanghai opens Apple Intelligence, now running on Alibaba's Qwen.
Production: licensed documentary photo, Vikshy grade
 Friday · 17 July 2026 · The daily ready reckoner for AI

Apple's AI in China will run on a homegrown Qwen model

Beijing cleared Apple Intelligence, built on Alibaba's Qwen. A marker of where AI's centre of gravity is shifting.

New modelOpenAI's GPT-5.6 brings frontier AI at a lower priceEducationUS teachers get a free year of premium ClaudeOpenFree open models keep winning developers on price
Read the 90-second brief ↓
The Daily Refresh · 90-second read

AI got more global this week

The whole day, distilled, every item ends where it should: with what it means for you.

1

Apple's AI in China will run on Alibaba's Qwen model

Why it matters: a mainstream product picks a Chinese model, and it shows governments help decide which AI you can use.

2

OpenAI's GPT-5.6 brings frontier AI at a lower price

Why it matters: the smartest tier now costs less, and a new agent can finish multi-step tasks for you.

3

US teachers get a free year of premium Claude

Why it matters: vetted AI reaches classrooms, a sign these tools are entering everyday life carefully.

4

Free open models keep winning developers on price

Why it matters: DeepSeek and Qwen pull in users worldwide, which pushes everyone's prices down.

You're caught up. See you tomorrow at 06:00 UTC. A quiet mind is the whole point.
Vikshy Flash · live

Small updates worth knowing

Transient but real, limit resets, brief outages, weekend promos. Verified sources only; the one surface that updates through the day.

Verified only what counts?
13:20DeepSeek's new V4 charges double during Beijing peak hours; off-peak prices stay the same.Access
10:05ChatGPT's brief global outage is resolved; the enterprise API and Codex stayed online throughout.Reliability
08:30Claude Sonnet 5 keeps its lower introductory price through 31 August.Value ▲
The Daily Frame · in depth
Product screenshot
Vikshy Grade
Frame 211
A developer opens the GPT-5.6 models on launch morning.
Production: licensed documentary photo · 9 Jul 2026 · browse the Frame archive →
In depth · Model Launch

OpenAI's GPT-5.6: the launch that reset the frontier race

Three models (Sol, Terra and Luna) now power ChatGPT and OpenAI's tools, and put OpenAI back on top.

T1 Primary3 min readFrame 211 · the Frame Archive →
What this means for you
👤EverydayIf you use ChatGPT, you are likely already talking to one of these new models. Everyday answers should feel a bit sharper.
💼At workWriters, coders and analysts get a stronger default helper. The mid-tier Terra aims to balance speed and quality.
🏢BusinessCompanies gain three price-and-power options to match models to jobs. Cheaper Luna suits high-volume, simple work.
🌍The worldThe launch keeps OpenAI at the front of a fast, three-way race. It raises the bar its rivals must meet.

Ledger entry: Release execution · Materiality 5 · T1 Primary → ▲ for OpenAI at the next refresh.

Today's stories

Plainly told, impact first

Every story carries the same promise, what happened, and what it means for you.

Documentary photo
Vikshy Grade
A product manager watches an AI agent draft a slide deck from a short brief.
Production: licensed documentary photo · 09 Jul 2026
Quick Take · Technology

OpenAI's new AI does multi-step work and costs less

GPT-5.6 tops a leading coding test, and a new agent finishes tasks across your apps.

T1 Primary2 min read
Documentary photo
Data · Vikshy
The Vikshy Score board on 17 July, with the top three labs within three points.
Data-as-image: the chart is the photo
Data · The Board

The top three AI labs sit within three points

OpenAI, Anthropic and Google are clustered at the top; challengers push from below.

T2 Measured90 sec
Documentary photo
Vikshy Grade
A developer in Jakarta runs a free open model on a laptop at a shared workspace.
Production: licensed documentary photo
Explainer · Global

Free Chinese AI models keep winning the world on price

DeepSeek and Qwen lead open models; low cost is pulling in developers everywhere.

T2 Measured2 min read
What moved · last 30 days

Every move, with its evidence

The score is only as trustworthy as its receipts. Here's exactly what changed each number this month.

✳
Anthropic ▲ +1.9
Claude Fable 5 still tops the independent intelligence index and US export controls were lifted. The score eased slightly as June's launch momentum settled, with Claude for Teachers adding adoption. T1 Primary T2 Measured T3 Reported Capability Acceptance Adoption
◍
OpenAI ▲ +1.4
GPT-5.6 and a top coding-agent score lifted OpenAI, but that launch bump is now easing. A brief July 15 outage weighed on its reliability standing. T1 Primary T2 Measured T3 Reported Release Capability Reliability
✕
xAI ▲ +1
Grok 4.5 launched with a strong coding focus, a No. 4 independent ranking, and low $2/$6 pricing. But xAI's acceptance and reliability standings stay low, easing its score. T1 Primary T2 Measured T3 Reported Capability Value Release
▲
Google DeepMind ▼ -0.4
Gemini 3.5 Pro slipped to a July 17 target after a full rebuild, and senior DeepMind researchers left for rivals. New low-priced media models offered some offset. T3 Reported T1 Primary Release Ecosystem Acceptance
◈
Alibaba (Qwen) ▲ +0.3
China cleared Apple Intelligence built on Alibaba's Qwen, a major distribution win, and Qwen is now the world's top open-model franchise. That lifted its adoption and ecosystem standing. T3 Reported T2 Measured Ecosystem Adoption Capability
❖
Mistral ▲ +0.3
Mistral launched its first robotics model and teased a new open-weight family. Its score eased modestly as faster-moving rivals widened the frontier gap. T1 Primary T3 Reported Release Ecosystem
⬦
DeepSeek ▼ -0.1
DeepSeek's V4 moved from preview to official release and it stays OpenRouter's top model near 20% of tokens. New peak-hour pricing slightly offsets its value edge. T2 Measured T3 Reported Adoption Release Value
–
1 provider broadly flat
Movement within rounding, no material change this month.
Plain AI · today's 60-second concept

What is an "AI agent"?

A normal chatbot answers one question at a time. An AI agent takes a goal and does the steps for you. It can search, write, and use apps like email or a spreadsheet. Think of it as a helper that can act, not just talk. Newer tools, like OpenAI's ChatGPT Work, aim to finish a whole task and hand you the result.

Full explainer + 40 more concepts →

Vikshy Pulse Gloomy Euphoric
62
Upbeat

Developers are debating coding agents and free Chinese models; the week's outage and trust questions keep some cautious.

Reception color from Reddit / HN / X, capped in the Score, never a rank input.

New here?

The Vikshy Score in 60 seconds, what it measures, what it ignores, and why a loud launch can't buy the #1 spot.

How the score works →