Gemini 4 Argon lands in Cursor, Devin, and Copilot with independent benchmark results
2026-10-13
Following Gemini 4 Argon's general availability on October 13 (per Google's Gemini API changelog), the model was integrated into Cursor, Cognition's Devin, and GitHub Copilot. Cursor reported Argon scoring 52.8% on its own CursorBench at default effort, on par with Claude Opus 5.5; Cognition reported 60.5% on its FrontierCode 1.1 benchmark and offered a 75% Devin discount through October 27, per Latent Space's AINews digest. Artificial Analysis reported Argon costs $2,407 to run its benchmark suite — cheaper than OpenAI's, Anthropic's, and Meta's latest models and close to GLM-5.3 — per Simon Willison, who also confirmed launch pricing of $2/$10 per million input/output tokens, doubling to $4/$20 after November 30, 2026, which would match Claude Opus 5.5's price and remain half of Claude Fable 5.1's and GPT-6 Astra's.
Significance 4: Independent benchmarking (Cursor's CursorBench, Cognition's FrontierCode, Artificial Analysis's cost data) corroborates Google's own coding-capability and cost claims and places the new model near parity with Claude Opus 5.5 at lower run cost — a lab-reported capability claim now independently reproduced, which the rubric treats as a significance driver, though SWE-bench Verified / Terminal-Bench numbers for this specific model aren't yet in hand.
Implications · machine-drafted, not owner judgment
If CursorBench and FrontierCode numbers hold up under wider testing, Gemini 4 Argon becomes a credible challenger on the coding-agent axis at a meaningfully lower per-task cost than Claude Opus 5.5, Claude Fable 5.1, or GPT-6 Astra — which matters because the current coding-agent pick (Claude Opus 4.8) predates all three of those newer models and hasn't been re-evaluated against any of them. This widens the gap between what the verdict file currently says and what's actually available, and raises the odds that a coding-agent refresh is overdue regardless of which model ultimately wins it.
- An independent SWE-bench Verified or Terminal-Bench run for Gemini 4 Argon specifically
- Whether Argon's coding edge holds after its November 30 price doubling
- A refreshed coding-agent evaluation that includes Opus 5.5, Fable 5.1, GPT-6 Astra, and Argon together
Sources
- primary Gemini 4 Argon generally available (GA) retrieved 2026-10-14
- curator [AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks retrieved 2026-10-14
- curator llm-gemini 0.35 retrieved 2026-10-14
Affected axes
Coding · Cost · Agentic / tool use