HomeAIAI ModelsOpenAI Staff Say GPT-6 Astra and 6.1 Sol Feel 50% Faster in...

OpenAI Staff Say GPT-6 Astra and 6.1 Sol Feel 50% Faster in ChatGPT

OpenAI staff say GPT-6 Astra and GPT-6.1 Sol should now feel about 50% faster for people using them through ChatGPT subscriptions, after a change to how the models are served rather than a new release. The claim reached the public through staff posts relayed on OpenAI’s developer forum, not through an announcement on openai.com.

On 5 October 2026, the OpenAI Developer Community documented Day 1 of a pledged 28 days of quality-of-life improvements (or a “full reset”). The thread cites staff posts saying GPT-6 Astra and GPT-6.1 Sol should feel about 50% faster by default for subscription users. [1]

That matters for how much weight the number can bear. It comes from staff channels relayed on the community forum, not from an official product blog or a published measurement.

What the thread says the change covers

As summarised in-thread, the speed-up is described as an inference optimisation (serving the existing models more efficiently), not a new model. It is said to apply across OpenAI products and partners that support Sign in with ChatGPT—examples named in the thread include OpenCode, Pi, Amp, and Devin. No user action is required; rollout was described as over a couple of hours. [1]

Community commenters noted that API speed is excluded, so API evaluations need not be re-run solely for this change. [1]

The GPT-6 speed figure comes from staff posts, not a benchmark

The “about 50% faster” figure is OpenAI staff’s own claim, as carried in the forum thread. User reports in-thread were mixed while the rollout was landing. [1]

It does not, on this evidence, mean a new GPT-6 release, a public benchmark suite, or an API-wide latency cut.

Open questions about the speed-up

  • Whether OpenAI will follow with an official product post that restates the claim.
  • How consistent the speed-up feels across devices, regions, and workloads once anecdotal reports settle.
  • Any precise measurement method behind the “~50%” figure. [1]

Update, 8 October 2026: Codex Auto-review made free

The “28 days” thread has moved on. On Day 2 (6 October), OpenAI’s Thibault Sottiaux posted on X that Codex Auto-review is now free for all users signed in through a ChatGPT account, switched on under Settings → Permissions → Auto-review [2]. The forum’s summary adds that it does not count against plan usage [3].

Auto-review is a second agent that decides on a Codex agent’s riskier requests, so the user does not have to approve every step. OpenAI’s documentation describes how it works but says it is “not a deterministic security guarantee” [4].

The source label is a little stronger this time. Day 1’s speed claim reached TSN through the forum. Day 2 comes directly from an OpenAI staff account, and the feature itself is documented on OpenAI’s developer site, though the “free” part is stated in the staff post rather than the documentation [2][4]. The forum thread has since been retitled with each new day; the link in source 1 still redirects to the same thread.

Evening, 8 October 2026: Day 4

Spotted via OpenAI Developers (@OpenAIDevs) on X and Thibault Sottiaux (@thsottiaux) on X.

Day 4 of the “28 days” push brought two changes, and they carry different labels.

Ultrafast for GPT-6.1 Sol. OpenAI’s changelog for 8 October lists “GPT-6.1 Sol Ultrafast in Codex and ChatGPT Work”. It is available on the Pro $500 plan and eligible Enterprise and Edu plans; for Enterprise it is “off by default”, and workspace owners must enable it [5]. In the API, OpenAI’s guide says Ultrafast “for GPT-6 Astra and GPT-6.1 Sol is available to all API users” [6]. It costs more. In Codex, Ultrafast “uses included subscription limits at 8x the Standard rate”, and purchased credits and Enterprise pay-as-you-go usage are billed “at 6x the Standard rate” [7]. OpenAI’s developer account describes it as “Near-Astra intelligence at up to 8x faster speeds than Sol Standard”. That speed figure is OpenAI’s own claim, not an independent measurement [8].

Instant steering. OpenAI’s Thibault Sottiaux posted: “We have improved steering to be instant, leading to the model now reacting much faster to adjustments you make”, so users can “course-correct direction in realtime” [9]. Steering here means redirecting the agent while it is working. TSN did not find this change in OpenAI’s changelog or documentation, so it rests on the staff post alone.

Earlier the same day, Sottiaux also posted a “Day 3 (encore)” note that OpenAI had “silently re-shipped codex cloud”, without details [10].

Also on 8 October: Codex CLI 0.162.0

Separately from the “28 days” posts, OpenAI published version 0.162.0 of its open-source Codex command-line tool at 19:55 BST. It adds tools for “creating and listing managed Git worktrees” (separate working copies of a project) when that feature is enabled, lets users “Pin tasks in the agent Command Center”, and includes sandbox fixes, among them keeping “ripgrep configuration from weakening deny-glob masks”. OpenAI lists these as bug fixes, not security vulnerabilities [11].

GPT-6 speed is now something OpenAI sells in tiers

Read together, the first four days of the push show OpenAI treating speed as a product in its own right. Day 1 was a free improvement: the same GPT-6 Astra and GPT-6.1 Sol models, served more efficiently, for everyone on a subscription that signs in with ChatGPT. Day 4 was a paid one: Ultrafast for GPT-6.1 Sol, which burns through Codex subscription limits at eight times the Standard rate. The first is a staff claim with no published method; the second is written into OpenAI’s changelog and documentation, with prices attached.

For people who spend their working day in ChatGPT or Codex, a faster default matters more than a benchmark win, because waiting on a model is what breaks concentration in long agent sessions. It also sits alongside OpenAI’s move to put GPT-6 into every ChatGPT tier, which brings more traffic onto the same serving systems, and its pitch of GPT-6.1 Sol as near-Astra quality at a fifth of the price. The exclusion of the API is the practical limit: developers who measure latency in their own applications have no reason to re-run tests on the strength of the Day 1 post. Until OpenAI publishes how it measured the 50% figure, it is best treated as a well-placed staff claim, while the documented Ultrafast option is the change that teams can actually budget for.

Sources

  1. OpenAI Community — Day 1 of 28 days of quality-of-life improvements or a full reset — 5 October 2026 — https://community.openai.com/t/day-1-28-days-of-quality-of-life-improvements-or-a-full-reset/1403525
  2. Thibault Sottiaux (@thsottiaux), “Day 2.1/ We have made Auto-review free for all users signed in through a ChatGPT account…,” post on X, 6 October 2026, 08:13 BST. https://x.com/thsottiaux/status/2107368734981517634
  3. OpenAI Developer Community, “Day 2: Auto-review is now free…,” post in the “28 days of Quality of Life improvements” thread, 6 October 2026, 08:42 BST. https://community.openai.com/t/free-auto-review-day-2-of-28-days-of-quality-of-life-improvements-or-a-full-reset/1403525
  4. OpenAI, “Auto-review,” Codex documentation, read 8 October 2026. https://developers.openai.com/codex/concepts/sandboxing/auto-review
  5. OpenAI, “GPT-6.1 Sol Ultrafast in Codex and ChatGPT Work”, ChatGPT/Codex changelog entry dated 8 October 2026. https://learn.chatgpt.com/docs/changelog
  6. OpenAI, “Ultrafast mode”, API documentation, read 8 October 2026. https://developers.openai.com/api/docs/guides/ultrafast-mode
  7. OpenAI, “Speed” (Codex agent configuration), Codex documentation, read 8 October 2026. https://developers.openai.com/codex/agent-configuration/speed
  8. OpenAI Developers (@OpenAIDevs), X post, 8 October 2026, 19:26 BST (company post; the “up to 8x faster” Sol figure is OpenAI’s claim; the Codex docs state the 8x speed figure for GPT-6 Astra). https://x.com/OpenAIDevs/status/2108262812489531498
  9. Thibault Sottiaux (@thsottiaux), “Day 4/ We have improved steering to be instant…”, X post, 8 October 2026, 20:15 BST (staff post; the steering change goes beyond OpenAI’s published changelog and docs). https://x.com/thsottiaux/status/2108275041276420573
  10. Thibault Sottiaux (@thsottiaux), “Day 3 (encore)/ We silently re-shipped codex cloud…”, X post, 8 October 2026, 07:38 BST (staff post; no details given). https://x.com/thsottiaux/status/2108084615349170480
  11. OpenAI, “0.162.0” (tag rust-v0.162.0), openai/codex release notes on GitHub, published 8 October 2026, 19:55 BST. https://github.com/openai/codex/releases/tag/rust-v0.162.0

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. WashU Makes Edible Fungal Protein From Waste CO2
    AI Science & Health

More in this category

Latest stories

  1. WashU Makes Edible Fungal Protein From Waste CO2
    AI Science & Health

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.