Claude Haiku 5.5
Anthropic's smallest and fastest current model, built for high-volume, latency-sensitive jobs like classification, extraction and routing. It is the fastest model in this list on Artificial Analysis and still clears 40 on the Intelligence Index.
AnthropicLanguage modelReleased October 2026Data checked October 18, 2026
This model is not in the current people's rankings: it is either new and still collecting votes, or ranked below the top 12. See the AI leaderboard for the models people rate highest.
On paper
Fixed-test rankings from the benchmarks tab: Speed #1, Value #2, Reasoning #12.
Strengths and weaknesses
- Fastest output in this list (235 tokens per second on Artificial Analysis)
- Very low price ($0.10 / $0.50 per million tokens) with the same 1M-token context window as the bigger Claude models
- Artificial Analysis Intelligence Index of 43.4 at $0.21 per task, the second-best value in this list
- Much lower ceiling on hard reasoning than the larger Claude models (44.4% on Humanity's Last Exam vs 61.4% for Opus 5.5)
- Released on October 7, 2026, so it is not yet on the LMArena text, writing or vision leaderboards
Best for
- High-volume simple tasks
- Fast chat and summaries
- Low-cost automations
Every score we have
Each bar compares this model with the best result any model on this page has on the same test.
- LMArena WebDev (coding)1,584.7Arena score 1584.7 (95% CI 1573.2 to 1596.2); 3,156 votes; entry "Claude Haiku 5.5 (High)"; position 32 of 144 on the full board
- LMArena Agent Arena0.015IPS score 0.0149 (95% CI 0.0055 to 0.0244); 23,660 sessions; entry "Claude Haiku 5.5 (High)"; position 26 of 52 on the full board
- Artificial Analysis Intelligence Index43.443.4 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 43), setting "Claude Haiku 5.5 (max)"
- Humanity's Last Exam (Artificial Analysis)44.4%44.4% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "Claude Haiku 5.5 (max)"
- Terminal-Bench 4.0 (Artificial Analysis run)32.8%32.8% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "Claude Haiku 5.5 (max)"
- Long-context reasoning (AA-LCR)82.7%82.7% on AA-LCR (long-context reasoning), setting "Claude Haiku 5.5 (max)"
- AA-Omniscience, accuracy minus hallucination10.710.7 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "Claude Haiku 5.5 (max)"
- Output speed, tokens per second235 t/s235 output tokens per second (AA median), row "Claude Haiku 5.5 (max)"
- Cost per Artificial Analysis test task$0.21$0.21 to run one AA Intelligence Index task, row "Claude Haiku 5.5 (max)"
- Artificial Analysis Coding Agent Index41.441.4 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Claude Code - Haiku 5.5 (xhigh)"
- ARC-AGI-2 puzzles61.3%61.25% on ARC-AGI-2 (semi-private set), best listed setting "Claude Haiku 5.5 (XHigh)", $0.08 per task
- ARC-AGI-3 puzzles0.5%0.45% on ARC-AGI-3 (semi-private, standard harness; "provider adapter" runs excluded), best listed setting "Claude Haiku 5.5 (Max)"
- Humanity's Last Exam (Scale AI)30.1%30.12% on Scale AI's Humanity's Last Exam leaderboard (plus or minus 1.83), entry "Claude Haiku 5.5"
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.