source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

GLM 4.5V's OpenRouter row dropped its expiration_date again — 2026-12-31 in the snapshot fetched 2026-09-08T06:00:04Z, null in the one fetched 2026-09-09T06:00:04Z (both committed in 84a6649), with pricing and context length identical in the two — so the registry's expiration-date rule (absent -> active) returns the derived status to active. It is the third flip of this one marker since 2026-08-29 (see the annotations on the 2026-08-29 and 2026-09-02 changes to this row), so it reads as an unstable removal marker rather than a reprieve: it changes only the status badge, and the date was still null in the 2026-09-14T12:31:37Z snapshot.

source

source

source

source

source

source

source

source

The OpenRouter row for z-ai/glm-4.7-flash no longer carries the expiration_date 2026-09-10 that was present in the preceding snapshot, so its feed-derived status changes from deprecated back to active. This matters because the current row is the source of the model's lifecycle status: the change record cited here records a live reversal of the deprecation signal, not a new model or a change to its identity.

source

source

source

source

source

source

source

source

source

source

source

source

Correction: the 2026-09-05 departure was a substitution, not a retirement. The same-fetch arrival qwen/qwen3.8-max-0902 carries canonical_slug qwen/qwen3.8-max-20260902, sharing the dated qwen/qwen3.8-max stem with the departed row's qwen/qwen3.8-max-20260803. The original departure line remains in append-only history; the changed feed applies this correction and names the arriving row.

source

source

source

source

source

source

source

source

On 2026-09-04 the OpenRouter row for z-ai/glm-4.7-flash gained an expiration_date of 2026-09-10 that the 2026-09-03 snapshot carried as null, so this site's expiration-date rule (absent -> active, in the past -> retired, within 365 days -> deprecated) now derives deprecated where it derived active. It matters as a genuine lifecycle event — this row's first recorded status flip, so it cannot be a metadata correction of a dropped date — because the stub model/z-ai-glm-4-7-flash renders deprecated through its feed-bound $status fact and its timeline carries the mechanically appended deprecated event dated 2026-09-04, with the same rule deriving retired once 2026-09-10 passes if that date holds — and on this source it is a date, not a commitment: nex-agi/nex-n2-mini and nex-agi/nex-n2-pro carried 2026-09-04 in the 2026-09-01 snapshot, carried 2026-09-08 from the 2026-09-02 snapshot onward, and still derive deprecated rather than retired today.

Follow-up to the annotation above (j-20260904-08): the same 2026-09-04 snapshot it read carries the successor row that annotation left unnamed, and the feed names it outright rather than leaving it to inference. The pointer row ~z-ai/glm-flash-latest — described in the feed as "This model always redirects to the latest model in the GLM Flash family" — carries alias_target: {"name":"Z.ai: GLM 5.3 Flash","slug":"z-ai/glm-5.3-flash"}, OpenRouter's own declaration that z-ai/glm-5.3-flash is the current GLM Flash model. z-ai/glm-5.3-flash reads active — expiration_date "2098-12-31", the far-future sentinel this site's status rule reads as active — at 1,310,720 tokens of context against this row's 202,752, and on price it cuts both ways: 25% dearer on input (a prompt price of 0.000000075 against this row's 0.00000006) and 37.5% cheaper on output (0.00000025 against this row's 0.0000004). The precedent annotation on this same source (j-20260902-02) went looking for a migration target and reported that none was identifiable from the feed; here one plainly is, so a reader migrating off this deprecated row has a successor to name.

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

On 2026-09-02 the OpenRouter row for z-ai/glm-4.5v again carries expiration_date: 2026-12-31, where the 2026-09-01 snapshot carried null, so this site's expiration-date rule (absent -> active, within 365 days -> deprecated) returns the row to the deprecated status it last held before the 2026-08-29 flip to active — which the annotation on that change (j-20260830-02) read as a metadata correction on an otherwise unchanged row, and which today's reappearance reverts. It matters as a provisional migration cue, not an outage: the model remains callable — it is still listed and both its endpoints (Z.AI, Novita) serve it as of 2026-09-02 (https://openrouter.ai/api/v1/models/z-ai/glm-4.5v/endpoints) — and OpenRouter's own models API schema (https://openrouter.ai/openapi.json, components.schemas.Model.properties.expiration_date) defines expiration_date as "the date after which the model may be removed", so 2026-12-31 is the scheduled-earliest removal date rather than a firm one; the marker has now moved twice in four days, and no successor row is identifiable from the feed, so any migration target has to be chosen from the current catalog rather than from this change.

source

source

source

source

source

source

The llm-releases description on this line — size, architecture and context undisclosed — was still what the feed's own page said on 2026-09-03, and that page likewise carried no benchmark table ('No benchmark scores recorded yet. Spotted some? Submit a correction.'), while the model card and config.json of thomsonreuters/Thomson-1.0-Small, the Hugging Face repo the release announces (both fetched 2026-09-03), disclose a 35B-total/3B-active mixture-of-experts repurposed from the open-weight Qwen3.6-35B-A3B, 262,144 tokens of native context (card 'Context Length', config.json text_config.max_position_embeddings), and a benchmark table whose Overall Avg. row reads 74.6 against 71.2 for Gemma 4-31B and 68.2 for Haiku 4.5. It matters because this line is the site's own record of the launch, and its 'undisclosed' understates the public artefact it points at — the correction in prose is the post on this site, 'Thomson Reuters says its $40M in-house model is on par with the latest frontier models', at /blog/thomson-reuters-thomson-model.

source

source

source

source

source

source

source

source

source

source

source

source

source

source

source

Nex-N2-Mini's OpenRouter row gained the 2026-09-04 expiration_date it did not carry in the 2026-08-31 snapshot, so the registry's expiration-date rule (absent -> active, within 365 days -> deprecated) now derives deprecated where it derived active — a genuine lifecycle event, unlike the two earlier status flips on this source, both of which were dropped expiration dates, not new deprecations. It matters: the stub model/nex-agi-nex-n2-mini presents deprecated (feed-bound $status) on its built page, its timeline carries the mechanically appended deprecated event dated 2026-09-01, and the same rule derives retired once 2026-09-04 passes.

source

Nex-N2-Pro's OpenRouter row gained an expiration_date of 2026-09-04 that it did not carry in the 2026-08-31 snapshot, so the registry's expiration-date rule (absent -> active, within 365 days -> deprecated) now derives deprecated where it previously derived active — a real lifecycle event: the row's first recorded status flip, so it cannot be a metadata correction of a dropped date. It matters: the stub model/nex-agi-nex-n2-pro renders deprecated on its built page through the feed-bound $status fact, its timeline carries the mechanically appended deprecated event dated 2026-09-01, and the same rule derives retired once 2026-09-04 passes.

source

source

source