Grok 4.6AUG 12
61AA Intelligence IndexFive points above Grok 4.5 at the same $2 input / $6 output list price per million tokens.
Independent research for AI buyers
How fast is your AI use case getting cheaper?
Tell us the workflow you are considering. We investigate its real cost curve, the forces moving it, and the moment the economics may cross in your favor.
Bespoke, fixed-scope engagements. No subscription. No off-the-shelf report.
ILLUSTRATIVE FIELD MODELADW—01
Cost per successful task
COST
YOUR BUY THRESHOLD
$1.00
$0.54
$0.31
$0.17
DECISION WINDOWQ3–Q4
20242025NOW+18 MO
Questions worth answering
When will voice agents be cheap and reliable enough for our support volume?How quickly is the cost of reviewing an insurance claim falling?Should we build document extraction now, or wait another model cycle?What will high-quality video localization cost at our scale in 18 months?
FRONTIER SHIFT · AUG 15, 2026
Two releases.
Two kinds of deflation.
Grok 4.6 shows capability rising while list price holds. Qwen3.8-Max shows a second frontier supplier matching that price with a larger context window and aggressive cache economics.
Read the full field noteQwen3.8-MaxNOW LIVE
1Mtoken contextA 2.4T MoE flagship at $2 / $6, with implicit cached input priced at $0.25 per million tokens.
THE ENGAGEMENT
One question.
Researched for you.
“AI is getting cheaper” is not a decision-useful answer. Your costs depend on your quality bar, throughput, human review, latency, integrations, and the market serving your exact workflow.
We define the unit that matters, gather primary and market evidence, normalize it for capability, and turn the result into a decision-ready brief.
01ADW
The cost curve
A capability-adjusted view of what your workflow costs to automate now—and how that price has been moving.
02ADW
The drivers
The models, vendors, infrastructure shifts, and workflow changes pushing the economics up or down.
03ADW
The scenarios
A range of plausible paths, with explicit assumptions instead of one deceptively precise forecast.
04ADW
The decision
A concise recommendation: buy, build, pilot, wait—or watch for a specific technical or price trigger.
OUR LENS
Price alone lies.
A cheaper model is irrelevant if it cannot complete the task. We compare cost at a defined level of intelligence, quality, or reliability—then trace how the frontier is moving.
Capability-adjustedevery comparison, every scenario
128 models mapped by task cost and intelligence—not token price alone.
HOW IT WORKS
From fuzzy question
to timed decision.
We pin down the unit.
A short scoping session turns “AI for claims” into a measurable workflow, quality bar, and decision horizon.
We build the evidence.
We combine benchmark data, vendor economics, technical change, expert input, and your operating constraints.
We show you the window.
You receive the research, underlying assumptions, scenario model, and a live readout focused on your decision.
WHAT THIS IS
You are not buying our model.
You are hiring our attention.
We do not sell a static report library, a recurring dashboard, or a black-box forecast. Each engagement starts with the question your team actually needs answered.
The work is scoped and paid before research begins. The output is made for your use case, your constraints, and your investment decision.
WHO IT IS FOR
Teams deciding
where to place a bet.
01Operators
Evaluating when an AI workflow will beat the current process on cost and quality.
02Investors
Testing whether a company’s margin or moat survives the next wave of deflation.
03Vendors
Mapping where their offer sits on the cost curve and what buyers will expect next.
DETAILS
Before you ask.
What do we receive?+
A decision brief, the underlying evidence and assumptions, a scenario model, and a live readout with your team. Exact format follows the question.
How long does a study take?+
Most tightly scoped questions can be answered in two to four weeks. We confirm timing, sources, and deliverables before the engagement begins.
Do you sell subscriptions or data access?+
No. We conduct commissioned research. If you need the same question refreshed later, we can scope a follow-up.
What makes a good first question?+
A specific workflow, a real decision, and a time horizon. If the question is still broad, the scoping conversation is designed to sharpen it.
COMMISSION A STUDY
Bring us the decision
you cannot price yet.
Send the use case, the decision in front of you, and when you need the answer. We’ll reply with a proposed scope.
Start with your question