Claude Opus 5.5
Anthropic's flagship for long-running agentic coding and knowledge work, and the model Anthropic suggests starting with for most jobs. It tops the Artificial Analysis Intelligence Index and the LMArena Code Arena, and shares first place on the LMArena text leaderboard.
AnthropicLanguage modelReleased September 2026Data checked October 18, 2026
Where people rank it
#1
Chat and everyday questions (tied), 8,300 votes
#1
Coding and building apps, 2,775 votes
#1
Writing (tied), 1,945 votes
#1
Reading images and documents (tied), 2,508 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 #1, Speed #9, Value #15.
Strengths and weaknesses
- Highest Artificial Analysis Intelligence Index score in this list (57.6) and the best Humanity's Last Exam result measured by Artificial Analysis (61.4%)
- First on the LMArena Code Arena (WebDev) and first on Terminal-Bench 4.0 when run in Claude Code (64.9%)
- First on LMArena creative writing and on the LMArena Agent Arena
- At its maximum effort setting it costs $5.98 per Artificial Analysis task, several times more than value picks like GPT-6.1 Sol
- Thinking cannot be switched off, and its output speed (97 tokens per second on Artificial Analysis) trails Sonnet 5.5 and Haiku 5.5
Best for
- Building websites and apps with Claude Code
- Long, multi-step projects
- Polished long-form writing
- Hard research questions
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)1,508.2Arena score 1508.2 (95% CI 1500.8 to 1515.5); 8,300 votes; entry "Claude Opus 5.5 (High)"; position 3 of 416 on the full board
- LMArena creative writing1,519.1Arena score 1519.1 (95% CI 1504.0 to 1534.1); 1,945 votes; entry "Claude Opus 5.5 (High)"; position 1 of 414 on the full board
- LMArena WebDev (coding)1,807.3Arena score 1807.3 (95% CI 1793.1 to 1821.5); 2,775 votes; entry "Claude Opus 5.5 (Max)"; position 1 of 144 on the full board
- LMArena vision1,292.7Arena score 1292.7 (95% CI 1280.0 to 1305.4); 2,508 votes; entry "claude-opus-5.5-high"; position 11 of 165 on the full board
- LMArena Agent Arena0.189IPS score 0.1888 (95% CI 0.1614 to 0.2161); 10,551 sessions; entry "Claude Opus 5.5 (Max)"; position 1 of 52 on the full board
- Artificial Analysis Intelligence Index57.657.6 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 58), setting "Claude Opus 5.5 (max with fallback)"
- Humanity's Last Exam (Artificial Analysis)61.4%61.4% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "Claude Opus 5.5 (max with fallback)"
- Terminal-Bench 4.0 (Artificial Analysis run)59.6%59.6% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "Claude Opus 5.5 (max with fallback)"
- Long-context reasoning (AA-LCR)84.7%84.7% on AA-LCR (long-context reasoning), setting "Claude Opus 5.5 (max with fallback)"
- MMMU-Pro, image understanding87.7%87.7% on MMMU-Pro (image understanding, AA), setting "Claude Opus 5.5 (max with fallback)"
- AA-Omniscience, accuracy minus hallucination46.446.4 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "Claude Opus 5.5 (max with fallback)"
- Output speed, tokens per second97 t/s97 output tokens per second (AA median), row "Claude Opus 5.5 (max with fallback)"
- Cost per Artificial Analysis test task$5.98$5.98 to run one AA Intelligence Index task, row "Claude Opus 5.5 (max with fallback)"
- Artificial Analysis Coding Agent Index6666.0 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Claude Code - Opus 5.5 (max)"
- Terminal-Bench 4.0 leaderboard64.8%64.8% resolved on Terminal-Bench 4.0 (plus or minus 3.1, 95% CI) with Claude Code at max effort
- ARC-AGI-2 puzzles93.3%93.33% on ARC-AGI-2 (semi-private set), best listed setting "Claude Opus 5.5 (High)", $0.41 per task
Compare with
- 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.
- Muse Spark 1.3Meta's mid-priced Muse model, trained for agentic workflows and tuned for competitive coding, available through the Meta Model API.
- GPT-6.1 SolOpenAI's mid-priced flagship, described by OpenAI as near-Astra performance at a lower cost.
- Muse WonderMeta's newest frontier model for coding, long-running agents and computer use, released on October 12, 2026 through the Meta Model API and Muse Code.