Claude Haiku 5.5 scores one point above Grok 4.7 at a fifteenth of the cost per task
Claude Haiku 5.5 scores 36 against Grok 4.7's 35 on Intelligence Index v5.0, at $0.41 a task against $6.18. List price is $0.10 / $0.50 against $2 / $6.
GTM Architect & Growth Operator · Now · 1 November 2026 · 4 min read
On this page
Claude Haiku 5.5 is the smallest model in Anthropic’s lineup. On Artificial Analysis Intelligence Index v5.0 it scores one point above Grok 4.7, and running the index cost a fifteenth as much.
The numbers
| Info | Claude Haiku 5.5 | Grok 4.7 |
|---|---|---|
| Index score, v5.0 | 36 (max) | 35 (high) |
| Cost per index task | $0.41 | $6.18 |
| List price, in / out per 1M | $0.10 / $0.50 | $2 / $6 |
| Context window | 1M | 500k |
| Speed | 235 tokens/s | 87 tokens/s (xhigh) |
Haiku’s list price is a twentieth of Grok’s on input and a twelfth on output. Past 100,000 prompt tokens it rises to $0.50 and $2.50, which is still a quarter of Grok’s input price and 42% of its output price.
Effort changes the picture, and not in the direction you would guess
| Model and effort | Index | Per index task |
|---|---|---|
| Claude Haiku 5.5, high | 30 | $0.16 |
| Claude Haiku 5.5, max | 36 | $0.41 |
| Grok 4.7, low | 31 | $2.92 |
| Grok 4.7, high | 35 | $6.18 |
| Grok 4.7, xhigh | 34 | $7.51 |
Haiku at high scores one point below Grok at low and costs $0.16 against $2.92, eighteen times less. Grok’s top setting is not its best. Xhigh costs $1.33 more a task than high and scores a point lower, so on this run more effort bought a bigger bill and a worse result.
Both models talk a lot
Haiku generated 300 million tokens on the index. Grok at xhigh generated 140 million, against a median near 70 million. Verbosity flatters a per-token price, because the job costs more than the sticker suggests, and that makes the per-task row the one to read. Haiku still wins it by a wide margin.
Where Grok 4.7 still has a case
Not on the score. It has the smaller context window, the higher price, and a place below the average of comparable models on the index.
What it has is the ecosystem around it. The xAI API offers server-side tools for web search, X search and code execution. Grok Build is xAI’s terminal coding agent, in beta since May. Grok Bot runs saved routines in the cloud, which I cover in what stops your agent. If your work lives on X, or you already pay for SuperGrok, that is a reason. Grok 4.7 Fast, at twice the token rates, is only available in Cursor and Grok Build.
Which one I would use
For classification, extraction and routing, Haiku 5.5, which is what Anthropic built it for. Test it on a hundred of your own examples first, because one index is one index.
For a workload that needs X search, or a team already inside Grok Build, Grok 4.7. For everything else I would need a reason to pay fifteen times as much per task for a point less.
What would change my mind
A task where Haiku’s verbosity bites: long outputs billed at $0.50 per million still add up at 300 million tokens. A second index run that moves either score by more than a point. And any pricing change on Grok 4.7, which lists at the same $2 / $6 that Grok 4.6 did in August.
The month’s other releases are in the October digest.
Questions people asked
Is Claude Haiku 5.5 better than Grok 4.7?
On Artificial Analysis Intelligence Index v5.0 it scores one point higher, 36 at max effort against 35 at high. It also costs a fifteenth as much to run the index: $0.41 a task against $6.18.
How much cheaper is Claude Haiku 5.5 than Grok 4.7?
On list price, $0.10 input and $0.50 output per million tokens against $2 and $6, up to 100,000 prompt tokens. Above that Haiku is $0.50 and $2.50, still a quarter of Grok’s input price.
Does Grok 4.7 have a larger context window than Claude Haiku 5.5?
No. Grok 4.7 has 500k tokens. Claude Haiku 5.5 has 1M.
What is Claude Haiku 5.5 for?
Anthropic describes it for high-volume, latency-sensitive tasks such as classification, extraction and routing. It is the fastest model in the current Claude lineup.