A Three-Tool Approach To AI Work In September 2026

🔍 Read the full analysis: A Three-Tool Approach To AI Work In September 2026 on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get privacy and security gear delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

The AI frontier compressed into a price curve during September 2026, according to an analysis published on 29 September by Thorsten Meyer: six frontier models now sit within about 20 points of each other on the Artificial Analysis Intelligence Index v4.3.x, while their cost per task differs by roughly 100x. The release of GPT-6.1 Sol the same day — scoring 51 points at a cost of $0.39 per task — anchors a recommended three-tool workflow in which Claude Opus 5.5 builds, GPT-6.1 Sol reviews, and a decision model called Jev handles high-volume routing judgments.

Meyer’s core observation is that the question for AI users has shifted from “which model is smartest?” to “which model clears my quality bar at the lowest cost per task?” On the Artificial Analysis index, Claude Opus 5.5 (released 22 September) leads with a score of 58 at max effort, at $5.98 per task — roughly 17 tasks per $100. At the other end, GPT-6 Luna (released 22 September) scores 37 at $0.07 per task, or about 1,429 tasks per $100. Between them sit Claude Sonnet 5.5 (56 points, $7.60 per task), Claude Fable 5.1 (53 points, $7.63), GPT-6 Astra (53 points, $3.26), and the newly launched GPT-6.1 Sol (51 points, $0.39).

Three findings stand out in the data, according to Meyer. First, Opus 5.5 outscores its more expensive sibling Fable 5.1 by 5 points while costing less per task. Second, Sonnet 5.5 at max effort costs more per task than Opus at max while scoring 2 points lower, which Meyer argues makes it a poor fit at that setting. Third, GPT-6.1 Sol costs about one-eighth of Astra and one-twentieth of Fable per task for a score only 1 to 2 points lower.

GPT-6.1 Sol launched at the same listed price as its week-old predecessor — $2 per million input tokens and $10 per million output tokens — and is unusually concise. Artificial Analysis data cited by Meyer shows the high setting used 25 million output tokens on the index, against a median of 82 million for comparable models. The trade-offs are real: at high and xhigh effort, Sol takes 57 to 69 seconds to produce a first token, making it unsuitable for interactive use, and Opus 5.5 still leads it by 5 points at xhigh.

Why Price Per Task Now Drives Model Choice

The 100x cost spread across near-equivalent scores changes how teams should budget AI work. Meyer reports that the effort setting moves the bill more than the choice of model: on Opus 5.5, going from xhigh to max adds 2 index points but 73% more cost per task; from medium to max, cost rises 4.46x for 7 points. The practical consequence is that a “review seat” becomes affordable — at $0.32 to $0.39 per task, running GPT-6.1 Sol as an independent check on every meaningful change is cheap enough to be routine.

Meyer also argues that a different model family reviewing Opus output is a stronger check than Opus reviewing itself. His four operating rules: effort is not capability; more effort cannot fill in missing requirements; a second model reading the same flawed spec is not an independent review; and passing tests is not approval to ship.

He cautions, however, that cheaper tokens do not mean cheaper work. In an illustrative example, he estimates that halving model price saves about 12.5% of real cost, and a single extra minute of human review erases the saving.

September’s Release Timeline and the Index Behind It

September 2026 saw a rapid succession of frontier releases: Claude Fable 5.1 on 1 September, GPT-6 Astra on 3 September, Claude Opus 5.5 on 22 September, GPT-6 Luna on 22 September, Claude Sonnet 5.5 on 28 September, and GPT-6.1 Sol on 29 September. All scores in the analysis come from the Artificial Analysis Intelligence Index v4.3.x unless otherwise noted, and Meyer repeatedly stresses that the index is a map of general capability, not a verdict on any specific workload — he recommends shadow-testing before switching models.

The recommended stack assigns Opus 5.5 at high effort (54 points, $1.82 per task) for features, APIs, multi-file work and refactors; Opus at xhigh (56 points, $3.46) for architecture, migrations and trust boundaries; GPT-6.1 Sol at high or xhigh for detail dives and review; Astra or Fable as paid second opinions only when Sol and Opus disagree; and Sonnet 5.5 at high (47 points, $1.08) plus Luna for scoped subtasks and bulk classification. Jev, a decision model that cannot write a sentence, takes over high-volume yes/no and routing judgments.

“In four weeks, the AI frontier stopped being a leaderboard and became a price curve.”

— Thorsten Meyer, ThorstenMeyerAI.com

Gaps in the Sol Data and Index Noise

Several points remain unresolved. Artificial Analysis has not yet published low or max effort settings for GPT-6.1 Sol, so its full cost-quality range is unknown. Meyer notes that one index point is inside the noise, meaning the 1-to-2-point gaps between Sol, Astra and Fable may not be meaningful. The index itself measures general capability and may not reflect any specific user’s workload. Additionally, the cost-saving estimate about cheaper tokens is explicitly described by Meyer as illustrative rather than measured, and real-world savings will depend on how much human review time a team actually spends.

Watching for Sol’s Missing Settings

The near-term developments to watch are the publication of GPT-6.1 Sol’s low and max effort settings by Artificial Analysis, which will complete its cost-quality picture, and any pricing response from competing labs given Sol’s aggressive per-task economics. Meyer advises readers to shadow-test the stack against their own workloads before switching, and to expect effort-setting pricing — the largest cost lever he identifies — to remain a focus of model releases going forward.

Key Questions

What is the three-tool approach described in the analysis?

Thorsten Meyer uses Claude Opus 5.5 at high or xhigh effort as the main builder for development work, GPT-6.1 Sol at high or xhigh for detail investigation and independent review at $0.32–$0.39 per task, and Jev, a decision model, for high-volume yes/no and routing judgments. Alternates like Astra, Fable, Sonnet 5.5 and Luna are used only for specific jobs.

Why is GPT-6.1 Sol’s launch considered significant?

Sol scores 51 on the Artificial Analysis index at xhigh — only 1 to 2 points below Astra and Fable — while costing roughly one-eighth of Astra and one-twentieth of Fable per task. According to Meyer, this makes routine review passes on every meaningful change economically viable.

What are the drawbacks of GPT-6.1 Sol?

At high and xhigh effort, Sol takes 57 to 69 seconds to produce its first token, so it is not suited to interactive use. Opus 5.5 still leads it by 5 points at xhigh, and Artificial Analysis has not yet published its low or max settings.

How much does the effort setting affect cost?

On Opus 5.5, moving from xhigh to max adds 2 index points and 73% more cost per task; from medium to max, cost rises 4.46x for 7 additional points. Meyer calls effort the real cost lever — more influential than the choice between most models.

Does a cheaper model automatically reduce overall costs?

Not necessarily. Meyer’s illustrative estimate is that halving model price saves only about 12.5% of real cost, which a single extra minute of human review can erase. He describes this example as illustrative, not measured.

Source: ThorstenMeyerAI.com