What Makes My September 2026 AI Workflow Tick

🔍 Read the full analysis: What Makes My September 2026 AI Workflow Tick on ThorstenMeyerAI.com

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

Get the latest gadgets 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

A workflow report published on ThorstenMeyerAI.com on 29 September 2026 lays out a practical answer to a changed AI market: with six frontier models clustered within about 20 index points while their cost per task differs by roughly 100x, author Thorsten Meyer now runs Claude Opus 5.5 as his main building model and the same-day-released GPT-6.1 Sol as a low-cost reviewer. The report reframes model selection around clearing a quality bar at the lowest cost per task rather than chasing the top of a leaderboard.

All capability scores in the report come from the Artificial Analysis Intelligence Index v4.3.x, which Meyer describes as “a map of general capability, not a verdict on your workload,” advising readers to shadow-test before switching models. At the top of the field, Opus 5.5 (released 22 September) scores 58 at its max effort setting at $5.98 per task, while at the bottom GPT-6 Luna scores 37 at $0.07 per task. Between them sit Claude Sonnet 5.5 (56 at $7.60), Claude Fable 5.1 (53 at $7.63), GPT-6 Astra (53 at $3.26) and GPT-6.1 Sol at xhigh (51 at $0.39).

Three findings anchor the report. 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 than Opus at max for 2 fewer points, which Meyer says makes it hard to justify 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 on 29 September at the same list price as its week-old predecessor — $2 per 1M input tokens and $10 per 1M output tokens. Even its medium setting matches the earlier GPT-6 Sol’s index score of 48 at one-fifth of that model’s $1.06 per-task cost. 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. Artificial Analysis has not yet published low or max settings for the model, and Meyer notes that one index point falls inside measurement noise.

Why Price Per Task Now Drives Model Choice

The report’s central argument is that the frontier has stopped being a leaderboard and become a price curve. When scores converge, the practical question shifts from “which model is smartest?” to “which model clears my quality bar at the lowest cost per task?” — a change with direct budget consequences for any team running AI at volume.

The effort setting emerges as the biggest cost lever. On Opus 5.5, moving from xhigh to max adds 2 index points for 73% more cost per task; from medium to max, cost rises 4.46x for 7 points. Meyer runs Opus at high (54 points for $1.82) for everyday development and reserves xhigh (56 points for $3.46) for hard problems such as architecture, migrations and trust boundaries, calling max “rarely worth it.”

cheap review is the second consequence. At $0.32 to $0.39 per task, a GPT-6.1 Sol review pass is affordable enough to run on every meaningful change, and Meyer argues a different model family reviewing Opus output is a stronger check than Opus reviewing itself. He also cautions that halving model price saves only about 12.5% of real cost in his illustrative example, since a single extra minute of human review can erase the saving — a figure he flags as illustrative rather than measured.

The September 2026 Model Release Wave

The report caps a crowded month: Claude Fable 5.1 launched 1 September, GPT-6 Astra on 3 September, GPT-6 Luna and Claude Opus 5.5 on 22 September, Claude Sonnet 5.5 on 28 September, and GPT-6.1 Sol on 29 September — the day the report was published. Four weeks of releases produced six models within about 20 index points of each other, the compression that motivated the report’s cost-first framing.

Meyer’s full stack assigns roles by cost and strength: Opus 5.5 at high for features, APIs and multi-file work; Opus at xhigh for architecture and migrations; Sol at high or xhigh for detail dives and independent review; Astra or Fable as paid second opinions only when Sol and Opus disagree; and Sonnet 5.5 at high (47 points for $1.08) plus Luna for scoped subtasks and bulk classification. A decision model called Jev, which cannot write a sentence, handles high-volume yes/no and routing judgements.

“In four weeks, the AI frontier stopped being a leaderboard and became a price curve. Six models now sit within about 20 index points of each other, while their cost per task differs by roughly 100x.”

— Thorsten Meyer, ThorstenMeyerAI.com

Limits of the Index and the Cost Figures

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 curve is unknown. Meyer notes that one index point is inside the noise, meaning several near-tied rankings in his table should not be read as strict orderings. The human-cost comparison — that halving model price saves 12.5% of real cost — is explicitly labelled illustrative, not measured. The report is also one practitioner’s workload assessment: index scores are general-capability maps, and results on other workloads, especially non-code tasks, are untested here.

Waiting on Sol’s Full Curve

The nearest development is Artificial Analysis publishing the missing low and max settings for GPT-6.1 Sol, which would show whether its cost advantage holds across the full effort range. Meyer says he will continue shadow-testing before any model switch and will keep Sol in the review seat while Opus handles building; whether Sonnet 5.5’s extreme output length at max (about 193k tokens per task, the most Artificial Analysis has measured) changes its value proposition at lower settings is another open question. The report itself will need revision if October releases compress the score gap further or reset the price curve.

Key Questions

What is the main model recommended in the report?

Claude Opus 5.5 at high or xhigh effort. High delivers 54 index points at $1.82 per task for everyday development; xhigh adds 2 points at $3.46 for hard problems like architecture and migrations. Max effort is described as rarely worth the cost.

Why use GPT-6.1 Sol for review instead of Opus itself?

According to the report, a different model family reviewing Opus’s output is a better check than Opus reviewing itself, and at $0.32 to $0.39 per task a Sol review pass is cheap enough to run on every meaningful change.

Do the index scores guarantee these results on other workloads?

No. Meyer states the Artificial Analysis Intelligence Index v4.3.x is “a map of general capability, not a verdict on your workload” and advises shadow-testing before switching models. One index point is also inside measurement noise.

What are GPT-6.1 Sol’s main drawbacks?

At high and xhigh effort it 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 index points at comparable settings. Low and max effort settings have not yet been published.

How much does the effort setting matter compared with model choice?

The report says choosing the effort level moves the bill more than choosing between most models. On Opus 5.5, going from medium to max raises cost 4.46x for 7 additional index points.

Source: ThorstenMeyerAI.com