How I Use Opus, Sol, And Jev For Different AI Tasks

🔍 Read the full analysis: How I Use Opus, Sol, And Jev For Different AI Tasks 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

The release of GPT-6.1 Sol on 29 September 2026 has crystallized a new reality in the AI market: six frontier models now sit within roughly 20 points of each other on the Artificial Analysis Intelligence Index, while their cost per task differs by about 100x. In a first-person analysis published the day of the launch, Thorsten Meyer of ThorstenMeyerAI.com lays out how practitioners can exploit that spread — running Claude Opus 5.5 as a main builder, GPT-6.1 Sol as a cheap second reviewer, and the non-generative decision model Jev for high-volume routing decisions.

According to Meyer’s figures, all drawn from the Artificial Analysis Intelligence Index v4.3.x, Opus 5.5 (released 22 September) posts the highest score at 58 on its max setting, at $5.98 per task — about 17 tasks per $100. GPT-6.1 Sol at xhigh scores 51 for $0.39 per task, roughly 256 tasks per $100, while GPT-6 Luna handles classification and routing at 37 points for $0.07 per task. Between those poles sit Sonnet 5.5 (56 points, $7.60 per task), Fable 5.1 (53, $7.63) and GPT-6 Astra (53, $3.26).

Three findings stand out in the analysis. 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 for 2 fewer points — Sonnet writes about 193k output tokens per task at that setting, the most Artificial Analysis has measured, according to Meyer. Third, Sol costs about one-eighth of Astra and one-twentieth of Fable per task for a score only 1 to 2 points lower.

Meyer reports that the effort setting is the biggest cost lever: on Opus 5.5, moving from xhigh to max adds 2 index points and 73% more cost per task, while medium-to-max raises cost 4.46x for 7 points. He runs Opus at high (54 points, $1.82) for everyday development and xhigh (56 points, $3.46) only for architecture, migrations and trust-boundary work. Sol’s high and xhigh settings carry a real penalty — 57 to 69 seconds to first token, per Artificial Analysis data cited in the piece — making it unsuitable for interactive use at those settings.

Why Cost Per Task Now Beats Leaderboards

The analysis argues the practical question for buyers has changed from “which model is smartest?” to “which model clears my quality bar at the lowest cost per task?” With scores compressed and prices diverging, choosing a mid-tier model at top effort can cost 8 to 20 times more per task than a slightly cheaper model for marginal capability gain.

The workflow Meyer describes also challenges single-vendor habits: a different model family reviewing the builder’s output is a stronger check than a model reviewing itself, and at $0.39 per task, a review pass becomes routine rather than exceptional. Conversely, he cautions that cheaper tokens do not mean cheaper 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 Month of Back-to-Back Frontier Releases

September 2026 saw a rapid succession of launches: Fable 5.1 on 1 September, GPT-6 Astra on 3 September, GPT-6 Luna and Opus 5.5 on 22 September, Sonnet 5.5 on 28 September, and GPT-6.1 Sol on 29 September — one day after its predecessor GPT-6 Sol, which scored 48 at $1.06 per task. Sol launched at the same published token pricing as GPT-6 Sol, $2 input and $10 output per 1M tokens, with per-task costs far below the $10/$50 tier of Fable and Astra.

Meyer’s stack assigns each model a role: Opus 5.5 high as main builder, Opus xhigh for hard problems, Sol high or xhigh for detail work and review, Astra or Fable only as tie-breaking second opinions when Sol and Opus disagree, and Sonnet 5.5 high (47 points, $1.08) plus Luna for scoped subtasks and bulk classification. Jev, a decision model that Meyer notes 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.”

— Thorsten Meyer, ThorstenMeyerAI.com

Benchmark Gaps and Noise in the Numbers

Several points remain incomplete or unverified. Artificial Analysis has not yet published low or max effort settings for GPT-6.1 Sol, and Meyer notes that a difference of one index point falls inside measurement noise — meaning Sol’s exact standing against Astra and Fable is not settled. The index itself measures general capability, not any specific workload, and Meyer’s cost-saving example is explicitly labeled illustrative rather than measured.

Whether Sol’s unusually low token output (25M on the index at high, versus a cited median of 82M for comparable models) holds up across real workloads, and whether its long time-to-first-token improves, are open questions. Jev’s role and performance are described only from the author’s own usage, without published benchmarks in the source material.

Watching Sol’s Full Curve and the Index v4.4

The next data points to watch are Artificial Analysis’s publication of Sol’s low and max effort settings, which will complete its cost-quality curve, and any speed improvements to its high-effort latency. Further model releases in October could compress the price gap again — or reopen it. For practitioners, the immediate step Meyer recommends is shadow-testing candidate models on real workloads before switching defaults, and reserving expensive top-effort settings for problems that demonstrably need them.

Key Questions

What is GPT-6.1 Sol and why did it launch one day after Sonnet 5.5?

GPT-6.1 Sol is an OpenAI-lineage frontier model released on 29 September 2026, one week after GPT-6 Sol and one day after Anthropic’s Sonnet 5.5. It launched at $2/$10 per 1M tokens and scores 51 on the Artificial Analysis index at xhigh, per the data cited by Thorsten Meyer.

Which model does Meyer recommend as a daily default?

He runs Claude Opus 5.5 at the high effort setting for everyday development (54 index points at $1.82 per task), reserving the xhigh setting for hard problems like architecture and migrations, and calling max effort ‘rarely worth it.’

Why use GPT-6.1 Sol instead of Opus for everything?

Cost. At $0.39 per task, Sol is cheap enough to run as a routine review pass on every meaningful change, and as a model from a different family it provides a more independent check than Opus reviewing itself. Its drawbacks are a 5-point lower score at xhigh and 57–69 second time-to-first-token at high effort settings.

What is Jev used for?

Jev is a decision model that cannot generate text. Meyer uses it for high-volume yes/no judgements and routing decisions, where generative capability is unnecessary and per-task cost matters most.

Are these index scores a reliable basis for choosing a model?

Not on their own. Meyer cautions that the Artificial Analysis Intelligence Index v4.3.x is a map of general capability, not a verdict on any specific workload, and that one-point differences fall within noise. He recommends shadow-testing on your own tasks before switching.

Source: ThorstenMeyerAI.com