GPT-6 Luna Ties GPT-6 Sol on Financial Scenario Analysis
GPT-6 Luna Ties GPT-6 Sol on Financial Scenario Analysis
On catalyst and scenario analysis, the closest thing to a trading judgment that the Fronset benchmark measures, GPT-6 Sol has the top score and GPT-6 Luna is statistically tied with it. In sync mode, GPT-6 Sol costs more than twenty times as much per run [^1].
The lesson is less about one model than about what price tells you here. The expensive models do not buy a lead, and the cheapest ones are not safe either.
What is measured, and how to read a tie
This task covers the analysis that feeds a financial decision, not the decision’s outcome. Nothing in this category measures returns, trade outcomes or market-data work, so no result here says a model makes better trades.
The findings list 21 models with scores on this task. Quality is a score out of 10 reported with an interval. The findings do not say what level the intervals represent, so they are used only to judge overlap. The rule is simple: when two models’ intervals overlap, treat them as tied and let price decide.
The scores and the prices
The 0.17-point gap between GPT-6 Sol and GPT-6 Luna sits inside overlapping intervals, so it is a tie, not a ranking [^1]. Batch mode narrows the price gap, at $0.02 against $0.0015, but does not close it [^1].
| Model | Score (out of 10) | Price per task run |
|---|---|---|
| GPT-6 Sol | 8.8 (±0.46) | $0.04 sync [^1] |
| GPT-6 Luna | 8.63 (±0.35) | $0.0017 sync [^1] |
| GLM-5.3 | 8.21 (±0.25) | $0.14 sync [^1] |
| Claude Opus 5.5 | 8.14 (±0.41) | $0.11 sync [^1] |
| Tencent Hy3 | 6.76 (±0.34) | $0.0012 [^1] |
| NVIDIA Nemotron-3 Nano 30B-A3B | 5.49 (±0.49) | $0.0012 [^1] |
The top of the table is one tied tier. GLM-5.3, Claude Opus 5.5 and Moonshot Kimi K3 all sit in it, at $0.10 to $0.14 per run, and none of them scores above GPT-6 Luna [^1]. GPT-6 Sol’s wide interval also means its lead over GLM-5.3 is not decisive.
The bottom of the table is the other half of the story. Tencent Hy3 and NVIDIA Nemotron-3 Nano 30B-A3B cost about what GPT-6 Luna costs and score far lower.
What to do with it
- Default to GPT-6 Luna. It is in the tied top tier at the lowest price there.
- Step up to GPT-6 Sol only for its higher point estimate. On these figures, that is all the extra spend buys.
- Do not pick by price alone. The findings recommend not using Tencent Hy3, the Nemotron models or Xiaomi MiMo V2.6 Flash on this work without review [^1].
Why the value pick is per task
GPT-6 Luna’s result does not carry to adjacent finance work. On weighing several perspectives into one decision, it scores 7.65 (±0.24) and is not in the leading group [^2]. That group overlaps heavily and includes cheaper options, such as GLM-5.3 Flash at 8.54 (±0.22) for $0.03 per run in sync mode [^2]. Some of those leaders’ figures come from separate comparisons that the findings do not relate, so they are best not ranked against each other.
GPT-6 Sol fares no better on that task, at 7.63 (±0.36), and the score covers only calls that came back: it answered on 63% of attempts [^2]. A pipeline that needs an answer on every call should weigh answer rate separately from score.
The method transfers between tasks; the pick does not. Find the tied tier on the specific task you run, then take the cheapest model in it.
The live benchmark at https://fronset.ai/benchmark/ lets you set the quality bar your financial task actually needs, compare batch against sync pricing, and check these figures yourself.