GPT-5.2 through GPT-5.5
Mar 2026
Research Lead working on coding capabilities and midtraining for the GPT-5.2 through GPT-5.5 series (Dec 2025 – Apr 2026)
Building something new — formerly OpenAI and SpaceX
Chris Koch
I train frontier AI models
I'm Chris Koch. I left OpenAI to build something new.
I was a research lead at OpenAI from 2023 to 2026, working on pretraining, midtraining, and AI coding. I contributed to GPT-4 Turbo, GPT-4o, GPT-4.5, the o-series, GPT-5, GPT-5.2 through GPT 5.5, and the gpt-oss open-weight models, and I shipped many of OpenAI's small models.
Before OpenAI, I was an early engineer on Starlink at SpaceX, where I worked on scaling the network 100× — from 10,000 users to over a million. I double-majored in Computer Science and Neuroscience at the University of Pennsylvania.
At OpenAI
2023 – 2026- PRETRAINCore contributor to big model pretraining and midtraining for GPT-4o, GPT-4.5, and GPT-5.2 through GPT-5.5
- MIDTRAINManager of the Midtraining team and its research direction
- SMALLShipped efficient, small models — GPT-4o Mini, GPT-4.1 Nano, GPT-5.4 Nano, and the GPT-4b Micro protein language model
- OPENResearch Lead for the gpt-oss open-weight models
- REASONCore contributor to the o1 and o3 reasoning models
- CODINGResearch lead for coding capabilities for GPT-5.2 through GPT-5.5 (Jan 2025 – Apr 2026)
Shipped
2023 – 2026// what I've shipped, newest first
GPT-5.4 Nano
Mar 2026
Research Lead, Pretraining · the smallest, fastest model in the GPT-5.4 line
GPT-5
Aug 2025
Contributor, Midtraining · OpenAI's unified flagship across reasoning and chat
gpt-oss-120b & 20b
Aug 2025
Research Lead, Pretraining · OpenAI's open-weight reasoning models, with worst-case frontier-risk evaluation
o3 & o4-mini
Apr 2025
Core Contributor · next-generation reasoning models
GPT-4.1
Apr 2025
Contributor · coding-focused API family with 1M-token context
GPT-4.1 Nano
Apr 2025
Creator, Research Lead · OpenAI's fastest, lowest-cost model — a 1M-token context window in a nano footprint
Next-gen audio models
Mar 2025
Contributor · gpt-4o-transcribe, gpt-4o-mini-transcribe, and the steerable text-to-speech model
GPT-4.5
Feb 2025
Contributor, Pretraining · OpenAI's largest pretrained model at release
GPT-4b Micro
Jan 2025
Advisor · a GPT-4o-derived protein-engineering model that redesigned the Yamanaka factors with Retro Biosciences — 50× more efficient stem-cell reprogramming
OpenAI o1
2024
Core Contributor · the first model trained to reason before answering
GPT-4o Refresh
Nov 2024
Contributor, Midtraining · the gpt-4o-2024-11-20 update — sharper creative writing, better file understanding, and 16K-token outputs
GPT-4o Mini
Jul 2024
Contributor, Pretraining / Midtraining · OpenAI's cost-efficient small model in the GPT-4o family
GPT-4o
May 2024
Core contributor, Pretraining · also contributed to GPT-4o Mini · OpenAI's omni flagship across text, audio, and vision
GPT-4 Turbo
Nov 2023
Contributor, Midtraining · faster, cheaper GPT-4 with a 128K-token context window
Agents & tool use
2023
Research · agentic capabilities and tool use, before the o-series and GPT-4o