Bottom line

Use Luna High by default.

And probably do not use Sol Max. General workloads favor Sol Medium for escalation. Coding gives Terra Max a much stronger high-end role.

General-purpose value chart

Higher means a stronger Artificial Analysis Intelligence Index score; farther left means a lower average benchmark task cost. The dashed line marks configurations that are not dominated on those two metrics. Green rings mark the recommended operating tiers.

Luna Terra Sol Recommended stack Strictly dominated
GPT-5.6 cost per task versus Artificial Analysis Intelligence Index Scatter plot comparing Luna, Terra, and Sol configurations. Recommended configurations are Luna Medium, Luna High, and Sol Medium.

A configuration is “dominated” when another tested option costs no more and scores at least as high. Frontier status is not automatically a recommendation: latency, token use, behavior, and workload fit remain outside this chart.

Choose the stack by workload

The general-purpose and coding benchmarks support different escalation paths.

General workloads

General-purpose routing

  1. Luna Medium — volume
    Routine work and high-volume subagents.
  2. Luna High — default
    Preferred general-purpose production tier.
  3. Sol Medium — escalation
    Harder analysis, architecture, planning, and difficult tasks.
Coding workloads

Coding routing

  1. Luna Medium — volume
    Tests, documentation, small fixes, and scoped implementation.
  2. Luna High — default
    The strongest general coding default in the fuller chart.
  3. Terra Max — escalation
    Difficult debugging, refactors, and complex implementation.

Luna Max is the premium-value intermediate step. Sol Max is the maximum-capability option.

General-purpose takeaways

Use the chart to eliminate clearly inefficient tiers, then apply workload evidence to the remaining choices.

Eliminate dominated tiers

These configurations are beaten on both displayed metrics or lack a score needed for comparison.

  • Luna None: same 5¢ cost as Luna Medium, but 11 index points lower.
  • Terra None: no reported index, so it cannot be evaluated.
  • Terra Low and Sol None: Luna High is cheaper and materially stronger.
  • Sol Low and Terra High: Luna Extra High is cheaper at the same score.
  • Terra Extra High and Max: beaten by cheaper, higher-scoring Sol tiers.

The economic center

Luna Medium owns the low-cost end. Luna High is the most compelling upgrade in the entire set.

  • Medium → High: +8 index for +4¢.
  • High → Extra High: +3 index for +5¢.
  • After that, every additional point gets progressively more expensive.

Raw value versus role

Luna Extra High, Luna Max, Sol High, Sol Extra High, and Sol Max remain on the two-metric frontier, but each asks you to pay more for smaller score gains.

  • Luna Extra High/Max: credible intermediate escalation steps.
  • Sol Medium: not the third-best point by raw index per dollar; it is the operational family-level escalation.
  • Sol High/Max: maximum capability only when marginal gains justify the steep cost.

Coding Changes the Model Mix

The newer Artificial Analysis Coding Agent Index view strengthens Luna as the default family while making Terra Max the practical high-end escalation.

Coding benchmark view

Approximate values transcribed from the supplied Artificial Analysis chart. They are intentionally shown as ranges or rounded figures rather than false precision.

Configuration Avg. coding task cost Coding Agent Index Operational role
Luna Medium~$0.50Below the attractive quadrantHigh-volume worker
Luna HighNear Terra MediumMaterially above Terra MediumPreferred coding default
Luna Max~$1.55~74–75Premium-value step
Terra Max~$2.75~77–78Difficult-task escalation
Sol Max~$7.10~80Maximum capability

Value tiers

Luna Medium is the inexpensive worker for subagents, tests, documentation, small fixes, and well-scoped implementation. It is a volume choice, not the hardest-repository-work default.

Luna High is the strongest general coding default: a meaningful score improvement over Luna Medium while remaining inexpensive, and more attractive than Terra Medium on measured score versus cost.

Luna Max is the premium-value step. It appears to beat Luna Extra High and competes strongly with higher Terra efforts at lower cost.

Capability tiers

Terra Max approaches Sol Max at far lower task cost. Use it for difficult debugging, larger refactors, unfamiliar repositories, complex implementation, and tasks that failed on Luna.

Sol Max is the strongest measured GPT-5.6 coding configuration, not the routine value choice. Reserve it for repeated failures, exceptionally difficult work, or cases where failure costs much more than inference.

The jump from Terra Max to Sol Max costs more than 2.5× for roughly two or three additional index points: sharp diminishing returns for maximum capability.

Coding configurations to de-emphasize
  • None and Low: Luna None, Luna Low, Terra None, and Terra Low deliver substantially weaker coding results.
  • Sol None: expensive relative to its measured coding score.
  • Luna Extra High: squeezed by Luna Max, which appears to cost roughly the same or less and score higher.
  • Terra Extra High: squeezed by Luna Max or Terra Max.
  • Terra Medium: not necessarily bad, but less attractive than Luna High on measured coding score and cost.
  • Terra High: reasonable, though Luna Max appears to be the stronger premium-value step.

Terra may still offer output-length or behavioral advantages that this two-axis view does not capture.

Benchmark limits

Use benchmark charts as directional evidence, then validate routing against real repositories and production tasks.

  • The Intelligence Index and Coding Agent Index use different task suites; their scores and absolute task costs should not be compared directly.
  • Benchmark score per dollar is not the same as successfully completed production tasks per dollar.
  • Output length, latency, retries, context growth, task failure, and downstream human review can change real-world economics.
  • Operational role specialization can justify a model that is not the next point on a raw benchmark frontier.
  • Intelligent routing uses the cheapest model that reliably completes the task, then escalates when evidence warrants it.

General-purpose source data

Sorted by average Artificial Analysis benchmark task cost. “Index / 1¢” is a simple raw benchmark ratio, not completed production tasks per dollar. “Pareto frontier” only means no tested configuration is both cheaper and equal or better on the Intelligence Index; it does not imply strong practical value.

Model Reasoning Cost / task Index Index / 1¢ Assessment

Sources

Benchmark data sourced from Artificial Analysis. Visit each model page for methodology, evaluation breakdowns, pricing, and current leaderboard context.

The Artificial Analysis Intelligence Index (v4.1) incorporates nine evaluations: GDPval-AA v2, τ³-Banking, Terminal-Bench v2.1, SciCode, Humanity’s Last Exam, GPQA Diamond, CritPt, AA-Omniscience, and AA-LCR. See Artificial Analysis methodology for details. Coding Agent Index figures on this page are approximate values from the newer supplied chart and are kept separate from the general Intelligence Index data.

Benchmark data, Intelligence Index, and Coding Agent Index from Artificial Analysis. Interpretation: cost and index are useful routing signals—not a substitute for workload-specific evaluation.