OpenAI makes some of the best models in the world, and its agent platform ambitions are serious: AgentKit brought a visual Agent Builder, a Connector Registry, and ChatKit for embedding agents; Workspace Agents succeeded custom GPTs for enterprises and plug into Slack and Salesforce; and Frontier, launched in February 2026, is the enterprise platform for managing "AI coworkers" with shared business context, permissions, and evaluation.
But 2026 also showed what building on a lab's platform means. In June, OpenAI announced it is winding down Agent Builder and its Evals product, with both gone from the platform on November 30, 2026, barely a year after launch, with teams pointed toward Frontier and code-level SDKs instead. That is the pattern with lab platforms: the tools orbit the lab's strategy, and the lab's strategy moves fast. If your revenue operations run on a builder that gets deprecated, "migrate by November" becomes your problem.
General Input is the counter-position: a neutral automation platform where OpenAI's models are one excellent option among two dozen, where workflows survive vendor pivots because they are not built inside any vendor's application, and where the operational layer, 500+ integrations, triggers, approvals, per-run costs, credential firewalling, is the product rather than a side quest.
What OpenAI does well
- The models. The GPT-5.6 family is exceptional, and General Input customers use these models heavily. This comparison is about the platform around them, not the models.
- Developer primitives. The Agents SDK, Guardrails, and ChatKit are strong building blocks for engineering teams shipping AI products.
- Enterprise ambition. Frontier's shared business context, execution environments, and permissions target real enterprise needs, for organizations ready to commit at that level.
- ChatGPT distribution. Workspace Agents meet hundreds of millions of users inside a product they already know.
How they compare
| General Input | OpenAI | |
|---|---|---|
| Business system integrationsGeneral Input: 500+ apps, 30,000+ operations, managed OAuth. OpenAI: a curated connector registry, far shorter. | 500+ | |
| Model choiceGeneral Input runs OpenAI, Anthropic, Google, xAI, and open-weights models side by side. OpenAI runs OpenAI. | ||
| Natural language workflow creation | ||
| No-code agent builderOpenAI is retiring Agent Builder on November 30, 2026. | ||
| Workflow triggers (webhooks, schedules, email) | ||
| Deterministic code workflows | ||
| Human-in-the-loop approval gates | ||
| Per-run cost transparencyGeneral Input prices every execution, broken down by step. OpenAI reports aggregate token usage. | ||
| Least-privilege credential scopingAgent Firewall limits each credential to specific operations, enforced outside the model. | ||
| Use your existing AI subscriptionsToken Tunnel runs your ChatGPT plan (and Claude, Gemini, Grok) inside workflows. | ||
| Self-hosted / on-prem deployment | ||
| Trains frontier models |
Where they differ
Platform stability vs. lab velocity
OpenAI ships fast and prunes fast. Custom GPTs gave way to Workspace Agents; Agent Builder and Evals launched in October 2025 and were marked for shutdown by June 2026; Frontier is the new center of gravity, until the next reorganization of the roadmap. None of this is a criticism of the models, but automation infrastructure has a different job than a research lab's product surface: your invoice-processing workflow should not need a migration plan every fiscal year. General Input's product is the automation layer itself, and workflows built two years ago still run.
One lab vs. the field
Model leadership has changed hands repeatedly, and pricing moves with every release cycle. A workflow built inside OpenAI's platform runs OpenAI models, permanently. The same workflow in General Input picks the best tool per job: GPT-5.6 for one step, Claude Opus 5 for deep analysis, Gemini for long context, an open-weights model like Kimi K3 or DeepSeek V4 where cost dominates, with auto-routing and per-step effort control if you would rather not choose by hand. When the leaderboard flips, you flip a setting, not your architecture.
The integration surface
OpenAI's Connector Registry covers the marquee names. General Input maintains 500+ integrations with 30,000+ documented operations behind one control plane, with managed OAuth and per-operation docs the agent reads at run time. Business automation lives and dies on the long tail: the regional accounting tool, the recruiting ATS, the industry-specific CRM. That tail is the catalog's job, and it is where lab platforms are thinnest.
Cost you can attribute
OpenAI's dashboard tells you what your org spent on tokens. It cannot tell you what the lead-enrichment workflow cost per lead last month. General Input meters every execution, broken down by model tokens, integration calls, and compute, priced in transparent credits ($25 buys 4,000). And if your team already pays for ChatGPT, Token Tunnel routes that subscription into your workflows, so the models you are already buying power the automation layer at no additional per-token cost.
When to use OpenAI
Build directly on OpenAI when you are shipping an AI product: engineers, code, SDKs, evals you run yourself, and a workload tuned tightly to GPT-5.6's strengths. If your company is enterprise-scale and all-in on OpenAI, Frontier is the intended home for that commitment.
When to use General Input
- You are automating operations, not shipping an AI product. Workflows connected to your CRM, billing, support, and data systems, built by describing them.
- Deprecation risk is unacceptable. Your automations should outlive any lab's quarterly platform strategy.
- You want the whole model market. Two dozen managed models from nine labs, plus your own keys, your own subscriptions, or local models on your own hardware.
- Finance and security have questions. Per-run cost attribution, approval gates, credential firewalling, audit logs, and on-prem deployment.
- Non-engineers build too. Plain-English and voice building for the team, CLI and API keys for the developers.
FAQs
Does General Input support GPT-5.6?
Yes. The GPT-5.6 family (Sol, Terra, Luna) and the GPT-5.4 line are in the managed catalog, and you can bring your own OpenAI API key or route your ChatGPT subscription through Token Tunnel.
We built agents in Agent Builder. What now?
Before the November 30, 2026 shutdown, rebuild the automations you rely on somewhere durable. Most Agent Builder flows translate to General Input by describing the same logic in the builder chat, and the result gains triggers, cost tracking, approvals, and integration reach in the process.
Is General Input's AI worse because it is not a lab?
General Input does not train models; it runs the labs' best ones. Your workflows use the same GPT-5.6, Claude 5, and Gemini 3 quality, with the freedom to mix them per step.
Better together
Most General Input customers are OpenAI customers too: GPT-5.6 powering workflow steps, chosen because it earned the job, inside a platform that also connects the other 500 systems, prices every run, and will not sunset your automation layer with a blog post.