My main driver
GPT-6 Astra
GPT-6 Astra is currently my main driver.
That is my personal choice. The figures below separately show how selected configurations perform in Artificial Analysis benchmarks.
Three more models on the cost frontier
A selection from the Artificial Analysis Intelligence Index v4.3. Each result belongs to a specific model version and reasoning effort.
Small budget
GPT-5.6 Luna (max)
37.50 index points · $0.18 per AA task
The cheapest entry in this selection. Its lower index score is a trade-off that needs to make sense for your task.
Model on Artificial AnalysisMore index points per task
GLM-5.3-Flash
41.91 index points · $0.25 per AA task
Scores above Luna max at about $0.25 per task here. A candidate to test on your own work.
Model on Artificial AnalysisThe highest score in this selection
Claude Fable 5.1 (max with fallback)
53.37 index points · $7.63 per AA task
Scores slightly above Astra max at more than twice the cost per task. The small score gap does not establish an advantage for your work.
Model on Artificial AnalysisQuality versus cost
13 selected configurations. A model is on the Pareto frontier when no other configuration in this selection scores at least as high and costs at most as much, with a strict improvement in at least one metric.
Cost per AA Index task (USD, logarithmic)
Frontier in this selection Outperformed by another configuration Selected
The cost axis is logarithmic: equal distances represent equal price ratios. The index axis starts at 34. All points use unrounded source values.
- Artificial Analysis Intelligence Index v4.3
- 52.81
- Cost per AA Index task
- $3.26
On the frontier of this selection
No other configuration shown improves either metric without making the other worse.
All 13 configurations and sources
| Model / configuration | Index | USD / AA task | Frontier |
|---|---|---|---|
| GPT-6 Astra (low) | 45.99 | $0.82 | Yes |
| GPT-6 Astra (medium) | 49.67 | $1.54 | Yes |
| GPT-6 Astra (high) | 51.05 | $1.72 | Yes |
| GPT-6 Astra (xhigh) | 52.51 | $2.31 | Yes |
| GPT-6 Astra (max) | 52.81 | $3.26 | Yes |
| GPT-5.6 Luna (max) | 37.50 | $0.18 | Yes |
| GLM-5.3-Flash | 41.91 | $0.25 | Yes |
| Claude Fable 5.1 (xhigh with fallback) | 53.18 | $5.98 | Yes |
| Claude Fable 5.1 (max with fallback) | 53.37 | $7.63 | Yes |
| GPT-5.6 Sol (high) | 42.50 | $0.81 | Yes |
| GPT-5.6 Sol (max) | 47.06 | $1.99 | No |
| DeepSeek V4.1 Flash (max) | 39.55 | $0.27 | No |
| Muse Spark 1.3 (max) | 48.17 | $1.60 | No |
What this comparison tells you
The index combines several benchmarks. Cost per task reflects their token usage and weighting, including reasoning and caching. It is not a quote for your prompts. The selection includes five Astra efforts, two Fable efforts, two Sol efforts, plus Luna, GLM, DeepSeek and Muse.
This frontier applies to this selection and these two metrics. Speed, local use, privacy and performance on your specific task are not included. Small score gaps may be within measurement uncertainty. Benchmark results are not personal experience reports.
AA also places all five Astra efforts on its broader cost/index frontier. AA analysis dated 9 September 2026
Data retrieved on . A fixed snapshot. No automatic live updates.
Archive: personal ranking from 22 August 2026
The earlier ranking remains here: Fable 5 in S+, Sol in A, and the separate Google row. It describes those earlier versions and my assessment at the time. My current main driver is shown above.