Artificial Analysis says Gemini 4 Argon (high) matches GPT-6 Astra (max) on its Intelligence Index and has a 15% hallucination rate, compared with 51% for Astra
Google's new Gemini 4 Argon equals GPT-6 Astra on the Artificial Analysis Intelligence Index at 60% of the Cost per Task with discounted prices
Context & Ripple Effects
Google had already put Argon in the hands of a small set of cybersecurity partners, positioning the release around coding, knowledge work and defense use cases rather than a broad rollout. Artificial Analysis had also placed GPT-6 Astra behind Opus 5.5 and Sonnet 5.5 on its earlier index, so parity with Astra is meaningful but not a claim of category leadership.
The comparison sharpens Google's commercial case by pairing its reported score with lower task cost and a lower reported hallucination rate. It also sits alongside a material caveat: employee concerns about real-world coding performance were disputed by Google, underscoring the difference between benchmark results and deployment evidence.
First-order effects
- Google gains an independent comparison it can use with its initial cybersecurity partners: Argon is reported to match GPT-6 Astra on the index at 60% of discounted cost per task, while posting a lower measured hallucination rate.
- Buyers evaluating Argon against Astra have a clearer cost-and-reliability screen, although the reported real-world coding concerns make direct workload testing important.
Second-order effects
- OpenAI faces a more pointed procurement comparison in accounts where task cost and error rates carry as much weight as a frontier-model label.
- Artificial Analysis' index gains influence as a purchasing reference because the result combines capability, cost per task and hallucination measurement rather than treating benchmark rank alone as decisive.
Third-order effects
- If comparable results hold across deployed workloads, frontier-model competition shifts toward cost per useful task and measured reliability, compressing the premium available for marginal benchmark leads.
- The gap between public benchmark performance and operational coding quality becomes a central diligence issue for enterprise AI buyers, favoring vendors that can substantiate both.
The trend: Frontier AI procurement is moving from choosing the highest-scoring model toward comparing capability, error rates and cost per useful task together.