Head to head
Claude Haiku 5.5 vs Gemini 3.8 Flash
Which is better, Claude Haiku 5.5 or Gemini 3.8 Flash?
Claude Haiku 5.5 finishes ahead of Gemini 3.8 Flash on the index, 87.9 to 87.9. It is the stronger of the two on Speed and Value, while Gemini 3.8 Flash still leads on Craft. Both are scored on the same rubric, in the same week, and neither placing is sponsored.
Updated
Claude Haiku 5.5
11 of 73
87.9
A small model priced like a rounding error that scores where mid-tier models did last quarter.
Gemini 3.8 Flash
09 of 73
87.9
Google's smartest Flash yet, eighth on the arena, and still free to start on.
The four scores, side by side
| Score | Claude Haiku 5.5 | Gemini 3.8 Flash | Difference |
|---|---|---|---|
| Craft | 85 | 88 | +3 |
| Speed | 95 | 90 | +5 |
| Control | 84 | 86 | +2 |
| Value | 95 | 88 | +7 |
| Index score | 87.9 | 87.9 | +0.0 |
Which to pick
Pick Claude Haiku 5.5 unless Craft is what decides it for you, which is exactly where Gemini 3.8 Flash is the better answer.
Claude Haiku 5.5
Haiku 5.5 is the cheapest model Anthropic has sold, at ten cents per million input tokens for normal-length prompts, and it scores above GPT-5.6 Terra on the Artificial Analysis index. It is built for classification, extraction, routing and the sub-agents a larger model hands work to. Prompts above 100,000 tokens cost five times as much, so long-document work is not where the bargain is. It is too new to have a human-preference rating yet.
Where it shines
- Ten cents per million input tokens for prompts up to 100,000 tokens
- Scores above mid-tier models from the previous generation
- Million-token window for the jobs that genuinely need it
Gemini 3.8 Flash
Gemini 3.8 Flash went generally available on 2 September and replaced the short-lived 3.7 Flash within weeks. It reads text, images, video, audio and PDFs, takes a million tokens of context, and sits eighth on the arena leaderboard, ahead of many full-price flagships. The free tier covers it, and the paid price is introductory until the end of the year, when it doubles. Its benchmark scores trail its human-preference rating, which is worth knowing before you trust it with long agent work.
Where it shines
- Eighth on the arena leaderboard at Flash prices
- Native video, audio and PDF input
- The free tier covers it, so prototyping costs nothing