GPT-5.5 — Model Card, Benchmarks, and How to Run It in TeamDay
Jozo· 7 min read· 2026/04/24
GPT-5.5OpenAICodexModel Cards2026Frontier AITeamDay

GPT-5.5 — Model Card, Benchmarks, and How to Run It in TeamDay

The Short Version

On April 23, 2026, OpenAI announced GPT-5.5 — codename Spud — and rolled it out to ChatGPT Plus, Pro, Business, and Enterprise users the same day. API access followed on April 24.

GPT-5.5 is the first OpenAI model that unifies text, image, audio, and video in a single architecture. No more stitching together GPT-5 + Whisper + Sora behind your agent. One model handles all modalities end-to-end.

For TeamDay, GPT-5.5 runs under the Codex harness and is now selectable on every agent.


Model Card

FieldValue
ProviderOpenAI
Model IDgpt-5.5 (and gpt-5.5-pro)
CodenameSpud
ReleasedApril 23, 2026 (API: Apr 24)
ModalitiesText, image, audio, video — in and out
Context window400,000 tokens
Tool useNative, parallel, computer use, browser
Harness in TeamDayCodex
Best forMultimodal missions, agentic computer use, knowledge work

What OpenAI Claims

The launch positioned GPT-5.5 as a step toward an "AI super app." Concrete capability claims:

  • Agentic coding — strongest gains over GPT-5 here
  • Computer use — operates software, fills forms, moves across tools until a task is finished
  • Knowledge work — researching online, analyzing data, creating documents and spreadsheets
  • Early scientific research — reading papers, replicating analyses, drafting follow-ups
  • Native multimodal — process and generate text, image, audio, and video in one call

The unified architecture is the part that changes how you build agents. With GPT-5, a "watch this video and summarize what was decided" task meant routing through a separate audio model. With GPT-5.5, it's one call.


How It Compares

ModelReleasedStrengthContext
GPT-5.5Apr 23, 2026Unified multimodal, computer use400K
GPT-5.5 ProApr 23, 2026+ extended reasoning400K
Claude Opus 4.7Apr 16, 2026Long-horizon coding, verification1M
Gemini 3.1 ProFeb 19, 2026ARC-AGI-2 reasoning2M

Opus 4.7 still leads on long-horizon coding and the largest context. GPT-5.5 leads on mixed-modality work: any mission that touches video, audio, screenshots, or "use this software" steps.


How to Use GPT-5.5 in TeamDay

How it's wired

GPT-5.5 reaches your TeamDay agents through the Codex harness — OpenAI's official agentic runtime. Selecting Codex as an agent's harness unlocks the entire OpenAI lineup — GPT-5, GPT-5.5, GPT-5.5 Pro — without us having to ship an integration update for each new model. The harness handles the model, the tool calls, and the computer-use steps natively.

1. Switch any agent to the Codex harness

Open the agent, go to Agent Settings → Harness → Codex, then choose gpt-5.5 or gpt-5.5-pro as the model. Same place as before — we just added the new model IDs.

2. Mix harnesses on one mission

A common pattern now: a Claude Code agent owns the planning and code review; a Codex agent owns the execution and computer-use steps; a Gemini CLI agent handles the giant-context document review. Configure per-step harnesses in Mission Templates → Steps.

3. Use it for video and audio missions

The unified architecture means GPT-5.5 can ingest a customer call recording, transcribe it, extract action items, and draft the follow-up email — in one mission, no separate Whisper or transcription tool. Try the Customer Call Triage mission template to see it.

4. Bring your own OpenAI key

Your contract, your rate limits. Drop the key into Workspace Settings → API Keys → OpenAI and TeamDay routes all GPT-5.5 calls through it.


When Not to Reach for GPT-5.5

  • Pure long-context coding — Opus 4.7's 1M context and verification behavior still win on multi-hour engineering missions.
  • Cheapest-per-token batch work — Sonnet 4.6 and Gemini 3.1 Pro are both cheaper for high-volume content loops.
  • Pure image or video generation — pair GPT-5.5 as the planner with gpt-image-2 or Seedance 2.0 as the executor.

TL;DR

GPT-5.5 is the new go-to inside TeamDay for mixed-modality missions — anything that crosses text, screenshots, audio, and video in one workflow. It runs under the Codex harness, plays well alongside Claude and Gemini agents, and is selectable on every agent today.

Hire your team · Browse the agent roster

Try 50+ AI Models Free — $20 in Credits

Access Claude, GPT-4.1, Gemini, Flux 2, Kling, Seedance, and more in one workspace. $20 free credits with a business email — no API keys, no credit card required.