Model review roundup benchmarks Gemini 4 Argon against GPT-6.1 Terra and Kimi K3.1 Pro

levithefirst @levithefirstOriginally posted 2026.10.28

Open original on X

this is coming a bit late but here's my review of the latest major AI model releases. →GPT-6.1 Terra - what OpenAI said it’s for fast visual reasoning, computer use, and daily tasks. drafts, forms, technical drawings, video edits. 1 → better than the last Terra? yes. Artificial Analysis 48 at max vs GPT-5.6 Terra at 42. cost per task $0.40 vs $1.40. 2 → rank in its field Artificial Analysis ranks it at 48, 4 points behind GPT-6.1 Sol (52), also behind Claude Opus 5.5 (58) and Sonnet 5.5 (56). 3 → strengths half the price of Sol with cached input $0.05. big drop in hallucination rate vs the old Terra (38% vs 88% on AA-Omniscience). live in ChatGPT Work, Codex, and the API. 4 → weaknesses not a new flagship. it is behind Sol, Opus, and Sonnet on intelligence. OpenAI best bet at hard coding is still Sol and Astra 5 → models still above it Claude Opus 5.5, Sonnet 5.5, and GPT-6.1 Sol. level with or behind Astra and Argon depending on effort. 6 → price per 1M tokens $1 in / $5 out. cached input $0.05. half of Sol. then for Kimi K3.1 Pro - what Moonshot said it’s for flagship for long-horizon agent work and coding. 2.8T MoE (~104B active), 1M context. 1 → better than the last Kimi? yes. Artificial Analysis 55 vs Kimi K3 at 44. price cut by a third. 2 → rank in its field Artificial Analysis 55. behind Claude Opus 5.5 (58) and Sonnet 5.5 (56). ahead of GPT-6 Astra and Gemini 4 Argon (~53). 3 → strengths same $2/$10 price as Sol and Sonnet. cost per AA task ~$2.11 vs Opus’s ~$6. strong agentic gains (Terminal-Bench ~67%). weights promised by Nov 4. 4 → weaknesses weights not out yet. slower than Opus on output. still trails the top two Anthropic models. 5 → models still above it Claude Opus 5.5 and Sonnet 5.5. Muse Wonder Full is also around 56. 6 → price per 1M tokens $2 in / $10 out. cached input $0.20. Highspeed tier is $4/$20.