Claude Haiku 4.5 vs Llama 4 Scout
Claude Haiku 4.5 (Anthropic, released October 2025) and Llama 4 Scout (Meta, released April 2025) sit at opposite ends of the price-performance curve. Haiku 4.5 is built to be Anthropic's fast, cheap-relative-to-Sonnet model with near-frontier reasoning. Scout is a mixture-of-experts model (17B active of 109B total params) built for cheap inference and enormous context. Both are available on OpenKey with the same key and the same flat 3% fee on top of provider list price.
Spec vs spec
| Spec | Claude Haiku 4.5 | Llama 4 Scout |
|---|---|---|
| Context window | 200K | 10M |
| Max output | 64K | 16K |
| Input modalities | text, image, file | text, image |
| Output modalities | text | text |
| Knowledge cutoff | — | Aug 31, 2024 |
| Released | Oct 15, 2025 | Apr 5, 2025 |
| Reasoning | optional | — |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
anthropic/claude-haiku-4.5
Input · 1M tokens
$1.00 + 3%$1.03
Output · 1M tokens
$5.00 + 3%$5.15
Cache read · 1M tokens
$0.100 + 3%$0.103
Cache write · 1M tokens
$1.25 + 3%$1.29
FEE — FLAT, EVERY MODEL3%
meta-llama/llama-4-scout
Input · 1M tokens
$0.100 + 3%$0.103
Output · 1M tokens
$0.300 + 3%$0.309
FEE — FLAT, EVERY MODEL3%
One workload, priced on both
10M input + 2M output tokens at each model's price, flat 3% fee included.
anthropic/claude-haiku-4.5
$20.60
$20.00 provider + 3%
meta-llama/llama-4-scoutCheaper
$1.65
$1.60 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| Claude Haiku 4.5 | Llama 4 Scout | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| Code | 1164 | #63 | 839 | #106 |
| Data viz | 1168 | #59 | 940 | #96 |
| Game dev | 1162 | #60 | 838 | #105 |
| UI components | 1155 | #60 | 824 | #100 |
| Websites | 1164 | #64 | 793 | #112 |
Head-to-head preference voting. How we filter and rank
Pricing math
Provider list price: Haiku 4.5 runs $1.00/M input and $5.00/M output tokens. Scout runs $0.10/M input and $0.30/M output — a 10x gap on input price. On OpenKey, add the flat 3% fee: Haiku 4.5 becomes $1.03/M input and $5.15/M output ($1.00 x 1.03, $5.00 x 1.03); Scout becomes $0.103/M input and $0.309/M output ($0.10 x 1.03, $0.30 x 1.03).
Run the numbers on a concrete workload: 10M input tokens + 2M output tokens. Haiku 4.5 costs $20.00 for that job; Scout costs $1.60. That's a 12.5x difference for the same token counts, driven mostly by the output price gap (5.00 vs 0.30 per million). If you're processing volume where quality differences don't matter much, Scout is the obvious cost play.
Coding and agentic performance
Haiku 4.5 posts a 43.9 coding index and 16.4 agentic index on Artificial Analysis; Scout posts 8.2 and 1.1 respectively. That's not a close race — Haiku 4.5 is roughly 5x ahead on coding and roughly 15x ahead on agentic tasks by these indices.
On Design Arena, Haiku 4.5 ranks 30th-64th across categories (asciiart, codecategories, dataviz, gamedev, svg, uicomponent, website, 3d) with Elo scores from 1083 to 1187. Scout ranks 96th-112th on its five shared categories (codecategories, dataviz, gamedev, uicomponent, website) with Elo scores from 793 to 940. Haiku 4.5 wins every category where both models are measured, often by 200+ Elo points.
Context and long-document work
This is where Scout flips the comparison. Scout's context window is 10,000,000 tokens; Haiku 4.5's is 200,000 tokens — a 50x gap (context ratio 0.02 favoring Scout). If your job is ingesting a huge codebase, a long transcript archive, or multi-document retrieval that doesn't fit in 200K tokens, Haiku 4.5 simply can't take the input; Scout can.
Max output differs too: Haiku 4.5 allows up to 64,000 completion tokens, Scout caps at 16,384. So for long input / short output workloads, Scout's context advantage matters more. For workloads that need long, structured output back, Haiku 4.5 has more room.
Modality and tooling
Haiku 4.5 accepts text, image, and file input and supports `reasoning` and `include_reasoning` parameters plus `top_k`, `structured_outputs`, and `tool_choice` — useful if you're building agent loops that need step-by-step reasoning traces. Scout accepts text and image input only (no file uploads) but supports a wider classic-sampling parameter set: `frequency_penalty`, `presence_penalty`, `repetition_penalty`, `min_p`, `logit_bias`, `seed` — useful for deterministic or fine-tuned sampling control. Neither model exposes a documented knowledge cutoff for Haiku 4.5; Scout's cutoff is 2024-08-31.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| Production coding agent | Claude Haiku 4.5 | 43.9 coding index vs 8.2, and native reasoning parameter support |
| Bulk document summarization at scale | Llama 4 Scout | $1.60 vs $20.00 for a 10M-input/2M-output workload |
| Ingesting a multi-million-token codebase or archive | Llama 4 Scout | 10M token context vs Haiku 4.5's 200K |
| Agentic multi-step tool use | Claude Haiku 4.5 | 16.4 agentic index vs 1.1 |
| Long structured output (reports, code files) | Claude Haiku 4.5 | 64,000 max completion tokens vs Scout's 16,384 |
| High-volume low-stakes classification | Llama 4 Scout | $0.103/M input token price, roughly 10x cheaper than Haiku 4.5's $1.03 |
Questions
- Which model is cheaper on OpenKey?
- Llama 4 Scout, by a wide margin. Provider price is $0.10/M input and $0.30/M output versus Haiku 4.5's $1.00/M input and $5.00/M output. With OpenKey's flat 3% fee that's $0.103/$0.309 for Scout versus $1.03/$5.15 for Haiku 4.5 — roughly 10x on input tokens.
- Which model has a bigger context window?
- Llama 4 Scout, at 10,000,000 tokens versus Haiku 4.5's 200,000 tokens — a 50x difference. If you need to process very large inputs in one call, Scout is the only option of the two.
- Which model scores better on coding benchmarks?
- Claude Haiku 4.5, with a coding index of 43.9 versus Scout's 8.2 on Artificial Analysis, and higher Elo across every shared Design Arena coding-adjacent category, including codecategories (1164 vs 839).
- What does a real workload cost on each?
- For 10M input tokens plus 2M output tokens, Haiku 4.5 costs $20.00 and Scout costs $1.60 at OpenKey pricing. The gap is driven mainly by output token price ($5.00/M vs $0.30/M).