GPT-5.6 Sol Usage: Tibo Resets Limits, ~18% Longer Quota, 5-Hour Cap Returns

OpenAI just admitted the Sol quota story out loud: plans were not nerfed — the model got hungrier.

On July 29, 2026, Codex/ChatGPT lead Tibo Sottiaux posted that OpenAI reset usage limits for all ChatGPT Work and Codex users, shipped efficiency fixes that should make typical GPT-5.6 Sol usage last about ~18% longer, and will restore the five-hour limit tomorrow after pausing it while investigating.

Primary: Tibo’s X thread. Prior context on explainx.ai: 5-hour cap temporary removal · 8M Work/Codex users · thinking-budget / banked resets.

TL;DR

What Tibo said (plain English)

Users reported Sol chewing Codex limits faster than expected. OpenAI’s investigation landed on behavior + product mechanics, not a silent plan downgrade:

1. Sol works longer — more tool calls, more subagent coordination, better at hard problems, worse for quota math.
2. Same effort label ≠ same burn — High on Sol can cost more tokens than High on GPT-5.5.
3. Programmatic tool calling / code mode — more flexibility to run tools in parallel or keep working while waiting → more responses per turn, more cached input, higher usage — especially on tool waits and many web searches.
4. Uneven impact — averages/medians looked fine at launch; the long tail of hard tasks was the miss.

Tibo’s closer is unusually direct for a limits thread: capability and efficiency did not improve at the same pace, and OpenAI should have recognized that sooner and been more upfront.

Why “we didn’t nerf plans” still felt like a nerf

From the user side, “same plan, less work done” is a nerf even when the rate card is unchanged. Sol’s agentic appetite raised the tokens per successful task for hard jobs. Median users who ask short questions were fine; power users running multi-hour Codex loops hit walls first — then posted screenshots. OpenAI’s Jul 29 package (reset + efficiency + timed return of 5h) is an attempt to separate communication failure from pricing change.

If you manage a team seat budget, translate Tibo’s points into policy:

The five-hour comeback — and why timelines hate it

explainx.ai tracked the temporary removal of the rolling 5-hour cap in mid-July as launch-week firefighting after Sol demand spiked (full guide). Tomorrow’s restore flips that pause back on.

Reply vibe on the announcement thread (paraphrased): please don’t — weekend grind vs family time, preference for self-set limits, “what’s the point.” Treat that as user signal, not OpenAI policy.

Rolling windows vs weekly pools

A 5-hour rolling window punishes bursty deep work more than a weekly pool with the same total tokens. That is why weekend warriors hate it: Saturday morning intensity burns the window before afternoon. Self-set limits (user-requested in replies) would let individuals trade burst for average — OpenAI restored the simpler ops lever instead.

Plan around Jul 30 as if the window is back unless OpenAI posts a delay. Finish long Sol sessions on Jul 29 after the reset while the pause still holds.

What “~18% longer” actually means

It is an expectation for typical use, not a published new rate card. Some accounts already see “significantly larger” gains from today’s efficiency work. Power users on Ultra-style agent loops, dense web search, or code-mode waits may still sit in the long tail — which is exactly the distribution OpenAI says it underweighted at launch.

If you are comparing models for cost, also read GPT-5.6 vs Claude Fable 5 and Cursor Router — routing cheaper models for easy turns is the practical hedge when Sol High is the burn.

How to measure the 18% on your account

1. Pick a fixed eval suite (10 coding tasks you ran last week).
2. Re-run after the Jul 29 efficiency deploy with the same effort settings.
3. Compare tasks completed per quota bar, not vibes.
4. Segment “easy autocomplete” vs “multi-tool agent” — the 18% is for typical, not your worst loop.
5. Log when code mode + search dominate; those are the long-tail burners Tibo named.

Also track thinking-budget / banked reset behavior from earlier July (Sol thinking budget post) so you don’t double-count relief from two different levers.

Practical checklist for today → tomorrow

Team playbook (Work seats)

• Publish an internal effort policy (when High is allowed).
• Prefer diff-scoped agent runs over whole-repo thrash.
• Keep a fallback model for boilerplate (Terra / open weights / Cursor Auto).
• Don’t schedule all engineers’ big refactors for the same Monday morning after weekly reset — stampeding herd burns shared org optics even when seats are individual.
• For Claude Code users on Sol via bridges, see Claudex setup so quota accounting isn’t split across UIs you forget to check.

How this fits the July limits soap opera

Same pattern: OpenAI patches quota experience in public while Sol’s agentic appetite stays the product. For parallel Claude/Cursor resets the same month, start at Claude / Codex / Cursor usage limits and Fable 5 + GPT-5.6 rate-limit resets.

Product lesson for agent builders

If your agent gets “smarter” by running longer, your metering UX becomes part of the model launch. Document appetite changes in the same post as capability wins — Tibo’s thread is the template other labs will be judged against.

Operator FAQ

Will the 18% fix my Ultra loops? Maybe partially — measure. Long-tail code-mode + search jobs were the miss; efficiency work helps typical use first.

Should I cancel Work? Usually no — reset + honesty is better than silent nerfs. Add a second vendor for burst weeks.

Is High always wrong? No. High is wrong when Medium solves the PR. Keep High for novel architecture and security-sensitive diffs.

What about ChatGPT Work vs Codex UI? Same Sol appetite; different surfaces. Track both bars (Work vs Codex guide).

Document your team’s effort policy the same day the 5-hour window returns so people do not invent folklore in Slack. Limits are part of the agent product now — treat them like rate cards, not surprises.

Slack announcement template (copy/paste)

OpenAI reset Work/Codex usage on Jul 29 and expects ~18% longer typical Sol runs after efficiency fixes. The rolling 5-hour limit returns Jul 30. Please: (1) finish long Sol jobs today, (2) default to Medium effort unless the task is hard, (3) audit code-mode + web-search loops, (4) keep a fallback model for boilerplate. We did not get a plan nerf — Sol got hungrier. Details: explainx.ai guide on Sol limits.

Pin it. Update it if OpenAI delays the window. Measurement > folklore. Add a calendar reminder for the morning of Jul 30 so nobody discovers the rolling window mid-refactor. If your org spans timezones, state whose “tomorrow” Tibo meant in UTC.

Closing note for explainx.ai readers

This launch moves fast; verify primary docs before you standardize tooling or brief a client. Re-check availability, pricing, and model IDs on the official pages linked above, then tell us what broke in production so we can update the guide. Follow @explainx_ai for follow-ups when the vendors ship the next patch — and prefer measured evals on your own traffic over screenshot economics. If you only needed a headline, you already have it; if you are implementing, the checklists above are the part that saves a weekend.

Related on explainx.ai

• ChatGPT drops 5-hour limit (temporary) — Jul 12
• Codex + Work hit 8M users
• Sol thinking budget / banked reset
• Claude / Codex / Cursor limit reset day
• Fable 5 + GPT-5.6 rate-limit resets
• ChatGPT Work vs Codex
• GPT-5.6 Sol in Claude Code (Claudex)
• Cursor Router auto model selection

Sources

• Tibo Sottiaux — Jul 29 Sol usage update

Quota language, 18% figure, and five-hour restore timing reflect Tibo’s July 29, 2026 public post. Confirm in-product meters and Help Center language after the Jul 30 restore.