Curated by StartupHub.ai
Databricks' AI Index: The state of enterprise AI tokenomics The results show why model rankings alone don't tell the full story. Enterprises need to evaluate model and harness combinations against their own workloads, using real usage data and traces to understand the trade-offs between performance and cost.
Rate
Quality vs Cost per task
75%Quality100%
GPT 6 Luna (claude code): Cost per task: 0.05, Quality: 89.
Claude Haiku 5.5 (claude code): Cost per task: 0.16, Quality: 85.
GPT 6.1 Terra (claude code): Cost per task: 0.18, Quality: 87.
GLM 5.3 Flash (claude code): Cost per task: 0.19, Quality: 89.
Claude Sonnet 5.5 (claude code): Cost per task: 0.32, Quality: 88.
Muse Spark Open (claude code): Cost per task: 0.4, Quality: 80.5.
GPT 6.1 Sol (codex): Cost per task: 0.6, Quality: 91.5.
DeepSeek V4.1 Flash 1017 (codex): Cost per task: 0.95, Quality: 93.
Muse Wonder (codex): Cost per task: 1.5, Quality: 95.
Claude Opus 5.5 (claude code): Cost per task: 1.6, Quality: 89.
Gemini 4 carbon (claude code): Cost per task: 2, Quality: 90.5.
GLM 5.3 (claude code): Cost per task: 2.2, Quality: 93.
GPT 6 Astra (claude code): Cost per task: 2.8, Quality: 92.
Gemini Argon 4 (codex): Cost per task: 5.2, Quality: 86.
Claude Fable 5.2 (claude code): Cost per task: 5.5, Quality: 92.5.
Claude Fable 5.1 (claude code): Cost per task: 7.5, Quality: 96.
Gemini 3.8 Flash (codex): Cost per task: 12, Quality: 83.
$0.1Cost per task$20
Cost per task read from the chart you shared. Quality read from the chart you shared.

