Gemini 4 Neon ranks third in Artificial Analysis testing
See how Gemini 4 Neon compares with Claude Opus 5.5 on Artificial Analysis coding scores and benchmark task costs.
Artificial Analysis reported on November 18 that Google’s Gemini 4 Neon scored 61 and ranked third on its Intelligence Index, behind GPT-6.2 Astra and Claude Fable 5.2, when tested at Neon’s xhigh effort setting.
For developers selecting a coding agent, the more relevant result is a separate comparison: Neon in Google’s Antigravity CLI scored close to Anthropic’s Claude Opus 5.5 in Claude Code while costing less per benchmark task. That makes Neon a candidate for workload-specific evaluation, rather than proof that it will be cheaper for every team.
The Intelligence Index is a composite, not a coding-only ranking. Its current version combines nine evaluations spanning software engineering, knowledge work, science and reasoning. Neon’s score was one point above Gemini 4 Carbon at high effort; the two models were tested at different effort settings.
In the separate Coding Agent Index, Artificial Analysis gave Neon at xhigh in Antigravity CLI a score of 59, compared with 58 for Opus 5.5 at max in Claude Code. The evaluator reported average costs of $6.86 and $12.61 per task, respectively—about 46% less for the Neon setup. Claude Fable 5.2 at xhigh in Claude Code also scored 58, at $15.47 per task. These comparisons cover model-and-agent combinations, so they do not isolate the models from their tooling.
The reported costs are averages for the evaluator’s tasks, not a price per successfully completed production change. Token consumption, caching, retries, task difficulty and the agent harness can change the bill for a team’s own repositories. Teams comparing the options should measure cost alongside completion quality on representative tasks.
Google’s November 18 launch announcement says Neon is rolling out on the paid Gemini API tier; Antigravity access requires a Google AI Pro or Ultra subscription. Artificial Analysis also notes that its comparison with Carbon uses Carbon’s standard pricing, not its discount available through November 30.
Comments ()