1 00:00:00,000 --> 00:00:05,280 Until this week, a fresh local Codex session could quietly switch off reasoning 2 00:00:05,360 --> 00:00:08,360 summaries before you typed a single prompt. 3 00:00:08,580 --> 00:00:11,920 A pull request merged on October 4th stops that. 4 00:00:12,034 --> 00:00:15,840 Thanks to Jellypod for helping make this daily show a reality. 5 00:00:15,968 --> 00:00:19,040 This is The Codex Changelog, I'm Ethan. 6 00:00:19,202 --> 00:00:19,902 And I'm Maya. 7 00:00:20,522 --> 00:00:27,081 The change is PR 50811, commit afb436d, merged into 8 00:00:27,161 --> 00:00:30,562 openai codex main on October 4th, 2026. 9 00:00:31,102 --> 00:00:33,562 It builds on something we mentioned briefly before. 10 00:00:34,122 --> 00:00:41,022 In 0.160.0, PR 49144 made the terminal UI preserve reasoning 11 00:00:41,122 --> 00:00:42,822 summary settings for server connections. 12 00:00:43,362 --> 00:00:45,142 That fixed connections to a server. 13 00:00:45,522 --> 00:00:49,742 This one closes the remaining gap, which is the embedded local TUI thread you get 14 00:00:49,762 --> 00:00:51,042 when you just run codex. 15 00:00:51,375 --> 00:00:52,335 Here's the mechanism. 16 00:00:52,479 --> 00:00:57,135 When you launched an embedded session without any explicit reasoning summary 17 00:00:57,192 --> 00:01:02,095 configuration, the TUI runner injected its own launch override. 18 00:01:02,223 --> 00:01:05,615 That override set model reasoning summary to false. 19 00:01:05,802 --> 00:01:10,495 So from the model's point of view, somebody had configured summaries off, 20 00:01:10,559 --> 00:01:11,855 even though nobody had. 21 00:01:12,035 --> 00:01:17,535 The model's own default, and any default coming from a remote or managed server, 22 00:01:17,595 --> 00:01:19,855 got overridden by a client that was guessing. 23 00:01:20,135 --> 00:01:21,995 And the models in question matter here. 24 00:01:22,415 --> 00:01:29,275 The brief names GPT 6.1 Sol, GPT 6 Luna, and Astra as models that may emit reasoning 25 00:01:29,495 --> 00:01:29,895 summaries. 26 00:01:30,475 --> 00:01:32,115 I'd stress the word may. 27 00:01:32,475 --> 00:01:35,595 This isn't a promise that every model will now show you a summary. 28 00:01:35,995 --> 00:01:39,735 It's that the decision goes back to the model and the server instead of being made 29 00:01:39,775 --> 00:01:41,415 by the TUI at startup. 30 00:01:41,792 --> 00:01:46,912 The fix is that embedded threads and remote or app server threads now share the same 31 00:01:46,958 --> 00:01:48,512 launch override resolution. 32 00:01:48,779 --> 00:01:53,632 If you haven't set a summary preference, the TUI sends nothing, 33 00:01:53,672 --> 00:01:57,072 and the upstream configuration and model defaults decide. 34 00:01:57,285 --> 00:02:00,352 If you have set one, it's forwarded faithfully. 35 00:02:00,464 --> 00:02:04,432 That applies to model_reasoning_summary, and also to 36 00:02:04,494 --> 00:02:08,752 features.concurrent_reasoning_summaries. 37 00:02:08,885 --> 00:02:14,192 Explicit false values are preserved, and explicit true values are preserved. 38 00:02:14,372 --> 00:02:17,792 The only thing that changed is the unconfigured case. 39 00:02:18,112 --> 00:02:19,592 Why does that matter in practice? 40 00:02:20,112 --> 00:02:20,792 Two reasons. 41 00:02:21,412 --> 00:02:26,452 First, if you run on a team where a server defines reasoning visibility defaults, 42 00:02:26,892 --> 00:02:32,012 maybe for auditability, your local TUI could have been undercutting that policy 43 00:02:32,052 --> 00:02:33,292 without anyone noticing. 44 00:02:33,992 --> 00:02:38,872 Second, if you ever wondered why summaries showed up in one surface and not in your 45 00:02:38,932 --> 00:02:41,632 terminal, this is a plausible explanation. 46 00:02:42,192 --> 00:02:46,492 I'd call that interpretation, not something the PR states, but the behavior it 47 00:02:46,532 --> 00:02:49,432 describes would produce exactly that symptom. 48 00:02:49,708 --> 00:02:50,748 So how do you use it? 49 00:02:50,888 --> 00:02:53,148 The zero config path first. 50 00:02:53,328 --> 00:02:55,948 Get onto a build that includes the change. 51 00:02:56,135 --> 00:03:02,908 It landed on main after the stable 0.160.0 release, and it should be in the 52 00:03:03,068 --> 00:03:08,988 0.162.0 alpha stream, around alpha 12 and alpha 13. 53 00:03:09,148 --> 00:03:12,668 Then launch codex with no summary settings anywhere. 54 00:03:12,855 --> 00:03:17,708 If your model or server defaults to producing summaries, they should now show up on 55 00:03:17,735 --> 00:03:18,348 their own. 56 00:03:18,775 --> 00:03:21,555 If you'd rather be explicit, you can say so in config. 57 00:03:21,955 --> 00:03:27,975 In your home codex config.toml, or in a project level dot codex config.toml, 58 00:03:28,535 --> 00:03:33,235 set model_reasoning_summary to true, or to false if you want them off. 59 00:03:33,895 --> 00:03:38,735 Then under the features table, set concurrent_reasoning_summaries to true if you 60 00:03:38,775 --> 00:03:41,255 want reasoning streamed alongside the response. 61 00:03:41,935 --> 00:03:43,275 There's also a command line route. 62 00:03:43,595 --> 00:03:47,555 You can pass the model reasoning summary startup flag with true or false, 63 00:03:48,015 --> 00:03:52,835 or put the value in a profile, and the runner now honors either boolean instead of 64 00:03:52,875 --> 00:03:54,435 treating unconfigured as off. 65 00:03:54,833 --> 00:03:58,753 Now the caveats, and the first one is the one most likely to bite. 66 00:03:59,046 --> 00:04:01,313 Stale workaround configs. 67 00:04:01,457 --> 00:04:08,433 Some people hardcoded model_reasoning_summary equals false in config.toml during 68 00:04:08,493 --> 00:04:15,393 earlier regressions, the brief mentions the 0.155.1 proxy choking problem. 69 00:04:15,537 --> 00:04:18,033 That line is now strictly respected. 70 00:04:18,193 --> 00:04:22,833 It was never wrong to have it, but it means you won't see the new default behavior. 71 00:04:23,046 --> 00:04:25,393 If you want it, delete the line. 72 00:04:25,777 --> 00:04:27,437 Second, stream overhead. 73 00:04:27,877 --> 00:04:31,857 Concurrent reasoning summaries send additional chunks over your connection. 74 00:04:32,477 --> 00:04:36,997 If you're behind a strict enterprise proxy that already struggles with long running 75 00:04:37,177 --> 00:04:42,657 HTTP streaming or chunked transfers, turning that on, or inheriting it from a server 76 00:04:42,697 --> 00:04:45,977 default, could mean latency or dropped connections. 77 00:04:46,457 --> 00:04:51,817 If that's you, test it before you roll it out, and keep the explicit false setting 78 00:04:51,937 --> 00:04:52,857 as your escape hatch. 79 00:04:53,250 --> 00:04:55,090 Third, channel timing. 80 00:04:55,154 --> 00:04:57,250 This is alpha stream material. 81 00:04:57,376 --> 00:05:04,290 Stable users on 0.160.0 won't see it until it reaches the next stable minor 82 00:05:04,350 --> 00:05:04,930 release. 83 00:05:05,143 --> 00:05:09,330 So if a teammate on stable says summaries look different from yours, 84 00:05:09,383 --> 00:05:11,570 check versions before you check config. 85 00:05:11,803 --> 00:05:13,843 A few smaller items from this week's commits. 86 00:05:14,343 --> 00:05:19,963 PR 50720, from October 3rd, fixes Shift plus Enter in Windows Terminal. 87 00:05:20,503 --> 00:05:24,643 Windows Terminal sends it as an escape sequence, ESC 13;2u. 88 00:05:25,203 --> 00:05:30,483 Codex now decodes that with a 50 millisecond deadline, so you get a newline in the 89 00:05:30,523 --> 00:05:30,963 composer. 90 00:05:31,523 --> 00:05:35,423 Before, keys could be dropped, stray Escape events could fire, 91 00:05:35,723 --> 00:05:37,303 or a draft could submit early. 92 00:05:37,542 --> 00:05:43,302 PR 50447, October 2nd, removes namespace_tools from 93 00:05:43,364 --> 00:05:44,662 ProviderCapabilities. 94 00:05:44,822 --> 00:05:50,502 Tool namespaces and collaboration tool configs no longer need provider gate checks, 95 00:05:50,622 --> 00:05:54,262 and they use Responses Lite serialization directly. 96 00:05:54,475 --> 00:05:57,862 If you maintain a custom provider, that's worth a look. 97 00:05:58,195 --> 00:06:04,835 PR 50384 adds an opt in for 16 KiB ARM64 code pages in macOS code 98 00:06:04,895 --> 00:06:05,215 signing. 99 00:06:05,775 --> 00:06:07,515 You enable it with the environment variable 100 00:06:07,775 --> 00:06:14,695 OAI_RCODESIGN_ARM64_16K_CODE_PAGES set 101 00:06:14,735 --> 00:06:15,215 to true. 102 00:06:15,975 --> 00:06:20,115 The stated purpose is supporting newer Apple Silicon kernel memory page alignments, 103 00:06:20,355 --> 00:06:22,895 so it mostly matters if you build Codex yourself. 104 00:06:23,167 --> 00:06:24,927 And two more from October 2nd. 105 00:06:25,274 --> 00:06:32,207 PR 50128 exposes current_turn_model through the API, so you can audit 106 00:06:32,260 --> 00:06:34,527 which model handled a turn in real time. 107 00:06:34,874 --> 00:06:41,567 PR 50082 lets V2 subagents inherit parent scoped tool registrations 108 00:06:41,627 --> 00:06:45,807 dynamically, so child agents pick up the tools the parent has. 109 00:06:46,113 --> 00:06:48,093 The action item for today is small. 110 00:06:48,533 --> 00:06:55,473 Open your config.toml, search for model_reasoning_summary, and decide whether that 111 00:06:55,513 --> 00:06:58,573 line is still a choice or just an old workaround. 112 00:06:59,293 --> 00:07:00,273 That's the show for today. 113 00:07:00,493 --> 00:07:02,733 We'll be back tomorrow with the next changelog.