Claude Haiku 5.5 vs Gemini 3.8 Flash

Where it leads

• Lower input and output prices, $0.10 and $0.50 vs $0.75 and $3.75 per 1M tokens
• Higher Arena WebDev score, 1589 Elo vs 1583 Elo
• Higher FrontierMath Tiers 1-3 (v2) score, 75.1% vs 68.4%
• Longer responses, up to 128,000 vs 65,536 output tokens
• Cheaper cached input, $0.01 vs $0.075 per 1M tokens
• Cheaper batch processing, $1 vs $7.50 for 10M input and 2M output tokens

Best for

• High-volume classification, extraction and routing under the 100,000-token line
• Cost-sensitive fan-out steps where a frontier model is overkill

Watch out for

• Priced by prompt length, and it is a cliff rather than a taper: a request whose prompt passes 100,000 tokens is billed at $0.50 input and $2.50 output for the whole request, five times the short rate
• Cache reads and cache writes both count toward that 100,000-token measurement, so a large cached prefix can tip a request into the higher band on its own
• It is the only current Claude model whose 1M window is not priced uniformly; every other 4.6-and-later model bills a 900k request at the same rate as a 9k one

Where it leads

• Higher Arena Text score, 1496 Elo vs 1425 Elo
• Higher GPQA Diamond score, 95.4% vs 89.6%
• Higher SimpleQA Verified score, 69.7% vs 23.8%
• Accepts audio and video input

Best for

• High-volume agentic and coding workloads where per-token cost dominates the decision
• Multimodal pipelines that need audio, video and PDF input in one model

Watch out for

• The listed price is promotional and ends 31 December 2026; on 1 January 2027 input and output both double, to $1.50 and $7.50, and cached input goes from $0.075 to $0.15
• Output pricing is billed inclusive of thinking tokens, so reasoning-heavy prompts cost more than the output length suggests
• Google’s published comparisons are against its own previous Flash release and were not independently reproduced here