Per SoC Status

Per-SoC status (RC1)

State of release candidate RC1 v2026.10.11-rc1 (OpenIMP agg-35 687a68e4, open-tx-isp 6e464ac7). "Fully open" = open driver and OpenIMP run from a flashed image. Evidence: the agg-35 soak ran on 10 cameras from 2026-10-10 18:04 to 2026-10-11 08:00 (about 14 h) with 0 kernel oopses, 0 unexplained reboots and 0 application restarts. Details: Changelog, Feature Matrix, Supported SoCs and Sensors. Release: v2026.10.11-rc1.

Release SoCs: T10, T20, T21, T23, T31. Experimental (not part of the release promise): T30, T40, T41.

SoC	Kernel	RC1 status
T10	3.10.14	Release. Fully open: H.264, hardware JPEG, second stream, OSD, motion detection, day/night, reload and boot guard. isp-m0 reports the analog gain step the AE really reaches (142 instead of 144 on the tested sensor), which a day/night rule needs; device-tested. Front-crop window kept until a downscaled channel opens. Audio input/output and AEC device-tested (ERLE 25 dB on a speech segment). Image controls partly documented. Seen in the agg-32 soak: a few rounds with Day mode in the dark and a black frame (3 events in 44 h).
T20	3.10.14	Release. Fully open: vendor-identical OEM rate controller by default, SMART/CBR/VBR, encoder ROI and chroma QP offset, OSD through the IPU hook, real motion detection, audio and AEC (ERLE 24 dB). isp-m0 gain cap fix for day/night. AeComp monotonic.
T21	3.10.14	Release. Fully open: vendor-identical eprc rate controller, reference-buffer sharing, lifted AE/ADR/defog/AWB, SetBrightness acts and sepia works (beyond vendor), ae_it_max_us acts, encoder ROI. Encoder denoise follows the vendor semantics. Audio and AEC (ERLE 20 dB; vendor stack on the same model 16 dB).
T23	3.10.14	Release. Fully open with the native Helix encoder, no vendor helper. Encoder ROI effective (capability marker OpenIMP_Cap_T23HelixRoi), front-crop hang fixed, AF statistics on by default, MSCA scratch buffer, vendor AE default. Rare Helix errno 5: 0 errors in 6 days since the kernel fix (provisionally closed). No real WDR. Known: one unexplained silent reboot of a camera under agg-34 (kernel-log capture installed); day/night flapping at dusk is decided by the application, the library and driver report sensor values.
T30	3.10.14	Experimental. No T30 camera in the lab: host and build evidence only. A source-only H.264 path exists (the T30 contributor's live gate is documented in docs/T30_STATUS.md). Matrix: 54 gaps of 347 vendor functions, all missing exports.
T31	3.10.14	Release, reference SoC. H.264, HEVC, hardware JPEG, OSD, rotation, AEC (ERLE 25 dB), encoder ROI (beyond vendor, up to 10 windows). Front crop is effective and guarded (refuses windows the MSCA would have to upscale). WDR statistics ring persistent across STREAMOFF (100 stop/start cycles, 20 reloads without anomaly). Known: one black picture of about 80 s during a day/night probe, not understood. AeComp works but the AE needs about 30 s to settle.
T40	4.4.94	Experimental. First OpenIMP device run (vendor kernel driver, vendor libimp side by side): AddSensor/EnableSensor, FrameSource 3840x2160 + 640x360, H.264 on both channels, OSD, IVS and more pass. Open: CBR delivers only about 20 % of the target rate, no H.265, GetBrightness returns -1. Matrix: 120 gaps of 398 vendor functions.
T41	4.4.94	Experimental. Fully open from a flashed image (H.264, H.265); reload verified (10/10); tuning (gamma, CCM, CSC presets, manual exposure readback) and driver controls (frame drop, sensor register, WDR enable, AF weight) device-tested; apitest 260 PASS / 9 FAIL. Open: channel 1 colour at night, night column noise, short IVS gaps; MSCA channel-1 latch (t41_msca_cfg_update=2, opt-in) is correct up to 768x432, larger sub-streams can hang the SoC. No audio hardware on the test board (API level only).

C100 is not covered.

Since RC1 (next, not released, no soak yet): T40 got frame-level rate control, H.265, ISP-drawn OSD and night-mode calls (device-tested on a T40 board with the vendor kernel driver); the open T40 kernel driver on the T40XP/sc830ai board delivers real frames but is in progress and not in next. T41 got seven ISP tuning calls (device-tested) and driver reload fixes (50 cycles clean). T31 CBR no longer collapses under an impossible scene (host test), and the crop upscale parameter is capped at 100 after a hang at 150 %. See Changelog.