Opus, Sol, And Jev At Work In My September AI Setup

🔍 Read the full analysis: Opus, Sol, And Jev At Work In My September AI Setup on ThorstenMeyerAI.com

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

Get audio and creator 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

GPT-6.1 Sol launched on 29 September 2026, and analyst Thorsten Meyer used its arrival to lay out a revised AI workflow in which Claude Opus 5.5 remains the main building model while the new, far cheaper Sol handles detail work and code review. The shift reflects a market that, according to Artificial Analysis index data cited in the piece, now has six models within roughly 20 index points of one another while their cost per task differs by about 100 times — turning model choice from a capability contest into a cost-per-quality calculation.

Meyer’s setup assigns clear roles. Opus 5.5 at high or xhigh effort is the primary model for features, APIs, multi-file refactors, and hard problems such as architecture and trust boundaries. GPT-6.1 Sol at high or xhigh serves as the second pair of eyes for deep dives into specific files and independent review passes. Sonnet 5.5, Astra, Fable, and Luna are alternates for scoped jobs rather than defaults, and Jev, a decision model that Meyer notes cannot write a sentence, takes over high-volume yes/no and routing judgements.

The numbers behind the split come from the Artificial Analysis Intelligence Index v4.3.x. Opus 5.5 scores 58 at max effort at $5.98 per task. GPT-6.1 Sol reaches 51 at xhigh for $0.39 — about one-eighth of GPT-6 Astra’s cost and one-twentieth of Claude Fable 5.1’s, for a score only 1 to 2 points lower. GPT-6 Luna sits at the cheap end at $0.07 per task for a score of 37. Meyer also flags that Opus 5.5 now outscores its more expensive sibling Fable 5.1 by 5 points while costing less per task, and that Sonnet 5.5 at max effort costs more per task than Opus at max for 2 fewer points.

Effort settings, not model choice, are described as the biggest cost lever. 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 4.46 times for 7 points. Meyer runs Opus at high (54 points, $1.82) for everyday development and xhigh only for hard problems. Sol has real limitations: at high and xhigh settings it takes 57 to 69 seconds to produce a first token, making it unsuitable for interactive use, and Artificial Analysis has not yet published its low or max settings.

Why Cheap Review Passes Change the Workflow

The arrangement matters because it makes a previously expensive safety practice routine. A review pass at $0.32 to $0.39 per task is cheap enough to run on every meaningful change, and Meyer argues that a different model family reviewing Opus’s output is a better check than Opus reviewing itself. For teams and individual developers facing similar pricing, the piece suggests the frontier question has moved from “which model is smartest?” to “which model clears my quality bar at the lowest cost per task?”

Meyer also cautions against reading cheap tokens as cheap work. Halving model price saves only 12.5% of real cost in his illustrative example, and a single extra minute of human review erases the saving — a reminder that model pricing is one input among several in total workflow cost.

A Month of Releases Compressed the Field

September 2026 produced a cluster of frontier releases that set up the current comparison: Claude Fable 5.1 on 1 September, GPT-6 Astra on 3 September, Claude Opus 5.5 and GPT-6 Luna on 22 September, Claude Sonnet 5.5 on 28 September, and GPT-6.1 Sol on 29 September. Sol launched at the same published prices as its week-old GPT-6 predecessor — $2 per million input tokens and $10 per million output tokens — with Astra and Fable priced at $10/$50 and Luna at $0.10/$0.50.

Notable measurements cited include Sonnet 5.5 at max effort writing about 193,000 output tokens per task, the most Artificial Analysis has measured, which drives its cost jump from $2.74 to $7.60. Sol is described as very concise: its high setting used 25 million output tokens on the index against a median of 82 million for comparable models.

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

— Thorsten Meyer

Score Gaps Within Noise, Settings Unpublished

Several points remain unsettled. Artificial Analysis has not yet published low or max settings for GPT-6.1 Sol, so its full cost-performance range is unknown. Meyer himself notes that one index point is inside the noise, meaning Sol’s 1-to-2-point deficit against Astra and Fable may not be a real capability difference. The index scores measure general capability and may not reflect any specific workload, which is why Meyer recommends shadow-testing before switching models. His cost-saving arithmetic on cheaper tokens is described as illustrative rather than measured.

Pending Index Data and Jev’s Undefined Role

Watch for Artificial Analysis to publish Sol’s remaining effort settings, which could shift its value calculation in either direction. Meyer’s stack will be tested by how well the Opus-builds, Sol-reviews split holds up under real workloads, and whether Jev’s routing role expands as decision-model usage grows. The broader open question is whether the September price compression continues — if a future release matches Opus-level scores at Sol-level prices, the entire role-based stack described here would need rethinking.

Key Questions

What is GPT-6.1 Sol, and when was it released?

GPT-6.1 Sol is a frontier model released on 29 September 2026, priced at $2 per million input tokens and $10 per million output tokens. In Meyer’s analysis it scores 51 on the Artificial Analysis Intelligence Index at xhigh effort, at $0.39 per task.

Why use Opus 5.5 instead of a cheaper model for building?

According to the Artificial Analysis Index v4.3.x cited in the piece, Opus 5.5 leads the field at 58 points at max effort and still beats GPT-6.1 Sol by 5 points at xhigh (56 versus 51). Meyer uses it at high effort for most development work and xhigh for hard problems, since max effort adds 73% cost for only 2 points.

What are GPT-6.1 Sol’s main drawbacks?

The article identifies two: 57 to 69 seconds to first token at high and xhigh settings, making it non-interactive, and a 5-point score deficit to Opus 5.5. Its low and max effort settings have not yet been published by Artificial Analysis.

What is Jev’s role in this setup?

Jev is described as a decision model that cannot write a sentence, used for high-volume yes/no judgements and routing tasks. It is one of six models in the stack but is not used for generation work.

Does a cheaper model automatically lower overall costs?

No. Meyer’s illustrative calculation shows that halving model price saves only 12.5% of real cost, and one extra minute of human review can erase that saving. He also warns that cheaper tokens do not reduce the cost of flawed specifications or failed reviews.

Source: ThorstenMeyerAI.com