Botond Ormandi

Hi

[Botond Ormandi @BotondOrmandi]

@TJeparskis i’m thinking about switching real hard too

[Botond Ormandi @BotondOrmandi]

@TokenGremlin OAI really did. They even stopped

[Botond Ormandi @BotondOrmandi]

@sama did we not see Opus 5.5 or what?

[Botond Ormandi @BotondOrmandi]

@Angaisb_ OAI is done. This is just brutal

[Botond Ormandi @BotondOrmandi]

@DannyIO @Tsufuruleader @OpenAI are u ok? this is work…

[Botond Ormandi @BotondOrmandi]

@SadAlbert10 they just did

[Token Gremlin @TokenGremlin]

I’ve spent the past few days gathering what I can about OpenAI’s plans around DevDay, combining information from people I trust with public signals across OpenAI docs, GitHub commits, API schemas, and recent product changes.

Here’s the current radar:

→ GPT-6 Sol
→ Agents API expansion
→ OpenAI-hosted sandboxes
→ persistent agent sessions
→ dynamic model / reasoning / service-tier switching
→ Secure MCP Tunnels
→ Vaults + environment secrets
→ WebMCP + Sites
→ Live / realtime voice infrastructure
→ webhooks, prompt caching, and context management

Some of these could land before DevDay, some during it, and others may simply be infrastructure for what comes next.

I’ll break down what each item actually means, where the signals are coming from, and why I think it matters in the comments so this post doesn’t become enormous.

But the bigger picture is already pretty clear:

OpenAI is building far more than just new models.

[Botond Ormandi @BotondOrmandi]

@Soso_fun_yt "While the model itself supports a broad context window, the application orchestrator aggressively pruned the conversation history, retaining user prompts but truncating earlier assistant responses down to approximately the last ten turns." what harness does this?

[Botond Ormandi @BotondOrmandi]

@bkpark @k_grajeda Yeah, chat does xhigh, but only on pro plans. And you are right, codex does compact the context, but what im saying is chat does too, it is just much less effective, and is hid from the user. I only tested this on my pro plan so the behaviour might be different on other plans.

[Botond Ormandi @BotondOrmandi]

@bkpark @k_grajeda it actually does, u just dont see it. I tested it with roughly 1.2m of tokens in, and 5.6 sol xhigh said it had experienced compaction roughly 20 times.

[Botond Ormandi @BotondOrmandi]

@ChatGPT shutting the servers down for marketing? smart.

[Botond Ormandi @BotondOrmandi]

@andraspii @claudeai Perplexity doesn’t even have their own models for this, lol

[Botond Ormandi @BotondOrmandi]

@shiri_shh It means Western labs offer subscriptions that are way better value than the API, so a lot of their usage never shows up here.

[Botond Ormandi @BotondOrmandi]

@caelin_sutch “Claude just rebuilt macOS from scratch in 8 minutes. Apple is cooked.”

[Botond Ormandi @BotondOrmandi]

@maria_rcks this goes so hard for an icon reveal

[Botond Ormandi @BotondOrmandi]

@MarcosHernanz Since 5.5, u cannot change it. It will change visually, but will not let you go above the standard context limit.

[Botond Ormandi @BotondOrmandi]

@techikansh Yeah I tried adding it manually too and got:
"The 'gpt-5.6-sol' model requires a newer version of Codex. Please upgrade to the latest app or CLI and try again." Works fine in the CLI though