Why My AI Stack Gives Opus, Sol, And Jev Different Jobs

🔍 Read the full analysis: Why My AI Stack Gives Opus, Sol, And Jev Different Jobs on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

GPT-6.1 Sol launched on 29 September 2026, and within hours it was slotted into a published working AI stack as a low-cost second reviewer alongside Claude Opus 5.5, the main builder model. The reasoning, laid out by Thorsten Meyer on ThorstenMeyerAI.com, is arithmetic rather than hype: six frontier models now sit within about 20 index points of one another on the Artificial Analysis Intelligence Index, while their cost per task differs by roughly 100 times, turning model selection from a leaderboard question into a price-and-role question.

According to Artificial Analysis Intelligence Index v4.3.x data cited in the piece, Opus 5.5 (released 22 September) scores highest at 58 on its top setting, costing $5.98 per task, while GPT-6.1 Sol at xhigh scores 51 for $0.39 per task and GPT-6 Luna scores 37 for $0.07 per task. Between those poles sit Sonnet 5.5 (56 points, $7.60 per task), Claude Fable 5.1 (53, $7.63), and GPT-6 Astra (53, $3.26). The author’s stated conclusion: Opus 5.5 for main development work, Sol for detail digging and review, and alternates like Astra, Fable, Sonnet 5.5 and Luna reserved for specific jobs rather than defaults. A fourth model, Jev — described as a decision model that cannot write a sentence — handles high-volume yes/no and routing judgements.

Three findings shape the stack. 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 the author says makes that setting hard to justify. Third, Sol costs roughly one-eighth of Astra and one-twentieth of Fable per task for a score only 1 to 2 points lower.

The article also argues that the effort setting is a bigger cost lever than model choice. 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. The author therefore runs Opus at high (54 points, $1.82 per task) for most development, reserving xhigh for architecture, migrations and trust-boundary problems.

Why Role Assignment Beats Model Loyalty

The piece matters because it documents a shift many practitioners are making: when scores converge, cost per task becomes the deciding variable. A review pass at $0.32 to $0.39 per task, as Sol offers, is cheap enough to run routinely on every meaningful change — something not viable at $3.26 or $7.63 per task. The author notes that a different model family reviewing Opus’s output is a better check than Opus reviewing itself, which is the core argument for a two-model stack rather than a single default.

The piece also carries a caution for readers: halving model price saves only about 12.5% of real cost, by the author’s illustrative estimate, and a single extra minute of human review can erase that saving. Cheaper tokens do not automatically mean cheaper work.

The Late-September 2026 Model Wave

The stack update follows a crowded four weeks of releases: Claude Fable 5.1 (1 September), GPT-6 Astra (3 September), GPT-6 Luna and Opus 5.5 (22 September), Claude Sonnet 5.5 (28 September), and GPT-6.1 Sol (29 September). Sol launched at the same published token prices as its week-old predecessor — $2 per million input tokens and $10 per million output tokens — with three effort levels already listed by Artificial Analysis.

On the index, Sol’s medium setting (48 points, $0.21 per task) matches the earlier GPT-6 Sol’s score at one-fifth of that model’s $1.06 cost per task, according to the cited data. Sol is also described as very concise, with the high setting using 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. Six models now sit within about 20 index points of each other, while their cost per task differs by roughly 100×.”

— Thorsten Meyer, ThorstenMeyerAI.com

Benchmark Limits and Missing Settings

Several limits are explicit. Artificial Analysis has not yet published Sol’s low or max settings, and the author notes that one index point is inside measurement noise — meaning Sol’s 1-to-2-point gap behind Astra and Fable may not be meaningful. The high and xhigh settings carry real tradeoffs: 57 to 69 seconds to first token, which rules Sol out for interactive use at those settings.

All scores come from a single benchmark source (Artificial Analysis Intelligence Index v4.3.x), and the author’s own cost-saving math on human review time is described as illustrative, not measured. The Jev decision model is named as part of the stack but no benchmark scores or pricing details are provided for it in the source material.

Watch Sol’s Full Curve and Max Settings

Watch for Artificial Analysis to publish Sol’s remaining effort settings (low and max), which the author says are not yet listed. The open practical questions are whether Sol’s 1-to-2-point deficit to Astra and Fable holds up outside benchmark conditions, and whether rivals respond on price now that a new model offers near-frontier scores at roughly one-eighth the per-task cost of its closest peers.

The author’s own guidance for readers is to shadow-test any candidate model on real workloads before switching, since the index is a map of general capability rather than a verdict on a specific workload.

Key Questions

What is the core argument of the AI stack described here?

With six frontier models scoring within about 20 index points of each other but varying roughly 100x in cost per task, the practical question shifts from ‘which model is smartest’ to ‘which model clears my quality bar at the lowest cost per task’ — leading to role assignment: Opus 5.5 builds, GPT-6.1 Sol reviews, and cheaper models handle bulk tasks.

Why was GPT-6.1 Sol given the review role?

According to the author, Sol scores within 1 to 2 points of Astra and Fable 5.1 on the Artificial Analysis index but costs $0.39 per task at xhigh instead of $3.26 or $7.63 — cheap enough to run a review pass on every meaningful change. A different model family also provides a more independent check than Opus reviewing itself.

What are GPT-6.1 Sol’s main drawbacks?

The cited data shows 57 to 69 seconds to first token at high and xhigh settings, ruling out interactive use, and Opus 5.5 still leads it by 5 index points at xhigh. Artificial Analysis has not yet published its low or max settings.

Why run Opus 5.5 at high instead of max effort?

On Opus 5.5, going from xhigh to max adds 2 index points for 73% more cost per task, and from medium to max cost rises 4.46x for 7 points. High delivers 54 points at $1.82 per task, which the author calls the best balance for most development work; max is described as rarely worth it.

Are these benchmark scores a reliable guide for choosing a model?

The author explicitly cautions that the Artificial Analysis Intelligence Index is a map of general capability, not a verdict on any specific workload, and recommends shadow-testing before switching models. One index point is also described as inside measurement noise.

Source: ThorstenMeyerAI.com