Claude Haiku 4.5 vs GPT-5 Mini
Claude Haiku 4.5 and GPT-5 Mini are both positioned as the fast, cheap tier of their respective families — Anthropic's answer to lightweight reasoning versus OpenAI's. Haiku 4.5 (released October 2025) leans on Claude's tokenizer and a 200K context window; GPT-5 Mini (released August 2025) doubles that to 400K context and 128K max output, with reasoning mode mandatory by default. The pricing gap between them is the real story here.
Spec vs spec
| Spec | Claude Haiku 4.5 | GPT-5 Mini |
|---|---|---|
| Context window | 200K | 400K |
| Max output | 64K | 128K |
| Input modalities | text, image, file | text, image, file |
| Output modalities | text | text |
| Knowledge cutoff | — | May 31, 2024 |
| Released | Oct 15, 2025 | Aug 7, 2025 |
| Reasoning | optional | always on |
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%
openai/gpt-5-mini
Input · 1M tokens
$0.250 + 3%$0.258
Output · 1M tokens
$2.00 + 3%$2.06
Cache read · 1M tokens
$0.025 + 3%$0.026
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%
openai/gpt-5-miniCheaper
$6.70
$6.50 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| Claude Haiku 4.5 | GPT-5 Mini | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1149 | #63 | 1114 | #73 |
| ASCII art | 1187 | #30 | 1173 | #33 |
| Code | 1164 | #63 | 1164 | #64 |
| Data viz | 1168 | #59 | 1167 | #62 |
| Game dev | 1162 | #60 | 1194 | #54 |
| SVG | 1083 | #56 | 1149 | #42 |
| UI components | 1155 | #60 | 1161 | #57 |
| Websites | 1164 | #64 | 1167 | #62 |
Head-to-head preference voting. How we filter and rank
Pricing math
Provider list price for Claude Haiku 4.5 is $1.00/M input and $5.00/M output; on OpenKey that's $1.03/M input and $5.15/M output after the flat 3% fee (price × 1.03). GPT-5 Mini lists at $0.25/M input and $2.00/M output, or $0.2575/M and $2.06/M on OpenKey. That's a 4x gap on input price. Run the numbers on a 10M-input / 2M-output workload and Claude Haiku 4.5 costs $20.00 while GPT-5 Mini costs $6.50 — GPT-5 Mini is roughly a third of the cost for the same token volume. If you're calling either model at scale, that difference compounds fast.
Coding and agentic benchmarks
Claude Haiku 4.5 has published Artificial Analysis scores: 29.6 intelligence index, 43.9 coding index, 16.4 agentic index. GPT-5 Mini doesn't have an Artificial Analysis score in this data set, so a direct index comparison isn't possible. On Design Arena's codecategories benchmark the two are close — Haiku 4.5 scores 1164 Elo (rank 63, 44.8% win rate) versus GPT-5 Mini's 1164 Elo (rank 64, 43.8% win rate). Effectively tied on that specific test.
Context and output limits
GPT-5 Mini's 400K context window is double Claude Haiku 4.5's 200K — a 0.5x context ratio in Haiku's favor when comparing Haiku to GPT-5 Mini. GPT-5 Mini also allows up to 128K max output tokens versus Haiku 4.5's 64K. For long-document work — large codebases, multi-file diffs, long transcripts — GPT-5 Mini's window gives more headroom before you need to chunk or summarize input.
Design Arena category breakdown
Across Design Arena's eight categories, Haiku 4.5 leads on 3D (1149 Elo, rank 63 vs GPT-5 Mini's 1114, rank 73), asciiart (1187 vs 1173), dataviz (1168 vs 1167), uicomponent (1155 vs 1161 — GPT-5 Mini actually edges ahead here), and website (1164 vs 1167, again GPT-5 Mini slightly ahead). GPT-5 Mini leads gamedev clearly (1194 vs 1162) and svg (1149 vs 1083, with a notably better rank of 42 vs 56). Neither model dominates every category — it splits by task type.
When to pick each
Choose GPT-5 Mini for cost-sensitive, high-volume workloads, longer-context tasks, and gamedev or SVG generation where it clearly outperforms. Choose Claude Haiku 4.5 for 3D and asciiart generation, or when you're already standing on Anthropic's tool-use and structured-output parameter set (it supports top_k and top_p, which GPT-5 Mini's parameter list doesn't include). Both models run on OpenKey with one API key, billed at provider list price plus a flat 3% fee.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| High-volume API calls on a budget | GPT-5 Mini | Costs $6.50 vs $20.00 on a 10M-in/2M-out workload |
| Long-document or large-codebase processing | GPT-5 Mini | 400K context window vs Haiku 4.5's 200K |
| 3D generation tasks | Claude Haiku 4.5 | 1149 Elo (rank 63) vs GPT-5 Mini's 1114 Elo (rank 73) on Design Arena |
| Game dev code generation | GPT-5 Mini | 1194 Elo (rank 54) vs Haiku 4.5's 1162 Elo (rank 60) |
| SVG generation | GPT-5 Mini | Rank 42 vs Haiku 4.5's rank 56 on Design Arena's SVG category |
| Large single-response output | GPT-5 Mini | 128K max completion tokens vs Haiku 4.5's 64K |
Questions
- Which is cheaper, Claude Haiku 4.5 or GPT-5 Mini?
- GPT-5 Mini is cheaper by a wide margin. On OpenKey, GPT-5 Mini runs $0.2575/M input and $2.06/M output, versus Claude Haiku 4.5 at $1.03/M input and $5.15/M output. On a 10M input / 2M output workload that's $6.50 for GPT-5 Mini against $20.00 for Haiku 4.5 — a 4x gap on input pricing alone.
- Which model has a bigger context window?
- GPT-5 Mini, at 400,000 tokens versus Claude Haiku 4.5's 200,000 — twice the context. GPT-5 Mini also supports up to 128,000 max completion tokens compared to Haiku 4.5's 64,000, which matters for long documents or large multi-file code tasks.
- Does Claude Haiku 4.5 or GPT-5 Mini score better on coding benchmarks?
- On Design Arena's codecategories test they're nearly identical: Haiku 4.5 scores 1164 Elo (rank 63, 44.8% win rate) and GPT-5 Mini scores 1164 Elo (rank 64, 43.8% win rate). Claude Haiku 4.5 also has a published Artificial Analysis coding index of 43.9, which isn't available for GPT-5 Mini in this comparison.
- Can I use both models through one API key?
- Yes. Both Claude Haiku 4.5 and GPT-5 Mini are available on OpenKey, which routes to 329 models across 52 labs through a single key, charging provider list price plus a flat 3% fee on each call.