AI Potluck
Model components / Fine-tuned / chat models

GPT-4.1

OpenAI

Previous-generation general-purpose instruction-tuned model (April 2025). Strong on HumanEval and real-world coding with 1M token context, but two numerical generations behind by mid-2026; OpenAI's current production line is GPT-5 → 5.2 → 5.4 → 5.5 (April 23, 2026), with GPT-5.5 Pro and 5.5 Instant variants. GPT-4.1 remains available via API and is still used in legacy integrations.

OpenAI GPT-4.1 (+ mini, nano), released Apr 14 2025 as API-first coding/instruct models with 1M-token context. Post-trained instruct (non-reasoning) model, closed/API-only; now behind the GPT-5 line. OpenAI's own announcement page answers HTTP 403 to a plain fetch, so the SWE-bench figure rests on the InfoQ and Helicone write-ups. Verified 2026-08-13 via OpenAI's model docs, TechCrunch and InfoQ.

Openness

1 high confidence
1.0
weights
closed
data
closed
code
closed
license
Proprietary(API-only)

A proprietary OpenAI API model with no weights, data or code released. OpenAI's own model documentation gives the exact context window, 1,047,576 tokens.

Adoption

4 low confidence
4.0

GPT-4.1 was an API-first developer/coding model (initially not in ChatGPT default), widely adopted as a cheaper GPT-4o replacement and integrated across tooling. No primary per-model usage count disclosed; level 4 reflects broad developer-platform distribution as a major OpenAI API model rather than a verified figure. Treat as directional.

Capability

3 high confidence
3.0

A solid non-reasoning coding model at its Apr 2025 release - 54.6% on SWE-bench Verified, 21 points above GPT-4o - but well below the 2026 reasoning frontier (o3 71.7, GPT-5 74.9, Opus 4.8 88.6). Mid-tier, measured against the frontier. OpenAI's own announcement page is not publicly readable, so the headline 54.6% is confirmed on InfoQ and Helicone instead, both of which report it against GPT-4o's 33.2%.

Verified 2026-08-13