Gemini 4 Argon
Google's flagship reasoning model, built for complex software engineering, math, science and long agent runs, with output of up to 1M tokens. It has the highest score on the LMArena text leaderboard and ties for first in creative writing, while sitting just behind the Anthropic and OpenAI leaders on Artificial Analysis.
GoogleLanguage modelReleased October 2026Data checked October 18, 2026
Where people rank it
#6
Coding and building apps (tied), 1,775 votes
From blind votes on LMArena, where people pick the better of two anonymous answers. See the full tables on the AI leaderboard.
On paper
Fixed-test rankings from the benchmarks tab: Reasoning #3, Value #7.
Strengths and weaknesses
- Highest score on the LMArena text leaderboard (statistically tied with Muse Wonder and Claude Opus 5.5)
- Tied for first on LMArena creative writing and 57.1% on Humanity's Last Exam (Artificial Analysis)
- 91.3% on ARC-AGI-2 and a 1,048,576-token context window shared by input and output
- The introductory price of $2 / $10 per million tokens rises to $4 / $20 on December 1, 2026
- New on the boards: fewer LMArena votes than older models, no LMArena Vision entry and no Artificial Analysis speed measurement yet
Best for
- The best-rated chat answers
- Writing and brainstorming
- Very long documents and very long outputs
Every score we have
Each bar compares this model with the best result any model on this page has on the same test.
- LMArena text arena (human votes)1,523.9Arena score 1523.9 (95% CI 1515.1 to 1532.7); 4,904 votes; entry "Gemini 4 Argon (High)"; position 1 of 416 on the full board
- LMArena creative writing1,515.6Arena score 1515.6 (95% CI 1496.1 to 1535.1); 1,057 votes; entry "Gemini 4 Argon (High)"; position 2 of 414 on the full board
- LMArena WebDev (coding)1,678.2Arena score 1678.2 (95% CI 1662.8 to 1693.5); 1,775 votes; entry "Gemini 4 Argon (High)"; position 9 of 144 on the full board
- LMArena Agent Arena0.132IPS score 0.1323 (95% CI 0.1190 to 0.1456); 30,489 sessions; entry "Gemini 4 Argon (High)"; position 5 of 52 on the full board
- Artificial Analysis Intelligence Index52.652.6 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 53), setting "Gemini 4 Argon (high)"
- Humanity's Last Exam (Artificial Analysis)57.1%57.1% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "Gemini 4 Argon (high)"
- Terminal-Bench 4.0 (Artificial Analysis run)57.1%57.1% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "Gemini 4 Argon (high)"
- Long-context reasoning (AA-LCR)79.7%79.7% on AA-LCR (long-context reasoning), setting "Gemini 4 Argon (high)"
- AA-Omniscience, accuracy minus hallucination42.442.4 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "Gemini 4 Argon (high)"
- Cost per Artificial Analysis test task$1.99$1.99 to run one AA Intelligence Index task, row "Gemini 4 Argon (high)"
- Artificial Analysis Coding Agent Index63.863.8 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Antigravity CLI - Gemini 4 Argon (high)"
- ARC-AGI-2 puzzles91.3%91.25% on ARC-AGI-2 (semi-private set), best listed setting "Gemini 4 Argon (High)", $0.32 per task
- ARC-AGI-3 puzzles49.8%49.79% on ARC-AGI-3 (semi-private, standard harness; "provider adapter" runs excluded), best listed setting "Gemini 4 Argon (Low)"
Compare with
- Claude Opus 5.5Anthropic's flagship for long-running agentic coding and knowledge work, and the model Anthropic suggests starting with for most jobs.
- Claude Sonnet 5.5Anthropic's mid-priced model, which Anthropic describes as its best mix of speed and intelligence.
- Claude Fable 5.1Anthropic's top tier for the hardest reasoning and long-horizon agent work, which Anthropic recommends when Opus 5.5 still falls short.
- GPT-6 AstraOpenAI's most capable model, aimed at complex reasoning, coding, research and document creation.