o4-mini
Language Deprecated Full comparison ↗ openai/o4-mini · by OpenAI
· decoder-only-transformer
Pricing — 1 offering(s)
About this offering: OpenAI's o4-mini reasoning model, released in April 2025, with a 200K-token context; reasoning tokens are billed as output. OpenAI lists a shutdown date of 2026-10-23 for o4-mini, with GPT-5.6 Terra as the recommended replacement (OpenAI deprecations page, checked 2026-10-11).
Input tokens
- $1.10 / 1M tokens (input) Current 2025-04-16 → 2026-10-22
Output tokens
- $4.40 / 1M tokens (output) Current 2025-04-16 → 2026-10-22
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how o4-mini fits into a cost-aware routing setup
See how →Capability profile
How o4-mini rates across core capability dimensions, with the task-level evidence behind each rating.
Core capability — strong multi-step reasoning at significantly lower cost than o3.
SWE-bench Verified (500 human-verified GitHub issues; patch must pass the repo's own tests; no file-location hints): resolves 68.1% — fails roughly 1 issue in 3. Vendor-reported (OpenAI, o3 / o4-mini launch; no custom scaffolding).
Reliable function calling; better integrated than earlier o-series models.
200K context.
Slower than GPT-4o due to reasoning budget; significantly faster than o3.
Coding benchmarks
View the full coding benchmark leaderboard →| Benchmark | Score | Harness | Source |
|---|---|---|---|
| swe-bench-verified ↗ | 68.1% 2025-04 | agentless | 🏢 vendor ↗ |
| aider-polyglot ↗ | 72.0% 2025-04 | aider | 📊 leaderboard ↗ |
| livecodebench ↗ | 72.8% 2025-04 | — | 📊 leaderboard ↗ |
Operator guidance
OpenAI lists o4-mini for shutdown on 2026-10-23, with GPT-5.6 Terra as the recommended replacement, so plan a migration rather than new work. Until then it's a lower-cost reasoning model than o3, at $1.10 / $4.40 per 1M tokens with reasoning tokens billed as output.
Use cases
- Complex reasoning, STEM problems, and algorithm design at mid-range cost
- Production agentic pipelines needing reasoning depth without o3's price
- Advanced coding assistance and code review
Limitations
- Reasoning tokens can multiply the effective output cost on complex problems
- Not suitable for simple, latency-sensitive tasks — use GPT-4o mini instead
- Capability ratings are qualitative, not a single cited benchmark run