Every call, timestamped.

Google

Detected by the catalog sweep.

OpenAI

Detected by the catalog sweep, confirmed on the next sweep 61s later — 371 alerts delivered.

Moonshot

Missed by the radar: no catalog we sweep listed it in time, so it was confirmed at the vendor's own page and fired by hand — 351 alerts delivered. The confirmation time is when that alert went out, not when the model launched.

Recursive

Verified live 2026-10-21 — Oct 21 · Recursive's own launch post (recursive.com/updates/introducing-recursive-flint) and model card (RecursiveSI/flint): a 39.8B-total / ~2.5B-active mixture-of-experts coding model built to run locally, Apache 2.0 weights on Hugging Face · vendor-reported 74.2% SWE-bench Verified, 89.7% SWE-bench Pro v2 and 54.9% Terminal-Bench 2.1 (read Oct 26) · context: the card documents a 262,144-token configuration, the post says up to 512k with 256k recommended · our sweep read it free on two gateways inside four hours: opencode/flint-40b-free (models.dev, Oct 21 18:24 UTC) and recursive/flint-v1.0:free (OpenRouter, 22:08 UTC), both $0/$0 per 1M at 262,144 ctx. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

MiniMax

Detected by the catalog sweep, confirmed on the next sweep 61s later — 17 alerts delivered. Late: OpenRouter listed the GA id at 03:12 UTC, about 8.7 hours before this alert. The hunt filed it as a variant of the Sep 27 preview, whose alert had already gone out, so nothing fired until a registry fix shipped. Since Oct 25 an id whose hunt is already spent pages the operator within a minute.

Inception

Verified live 2026-10-19. Missed. OpenRouter listed inception/mercury-3 at 13:45 UTC on Oct 16, three days before Inception's launch post, and no entry matched it. Since Oct 24 the radar reads the labs' own model lists without a key (a public mirror of them, plus the labs' own docs and pricing pages) and emails the operator within a minute about any id there that no hunt covers, with a one-tap way to send the alert.

Reflection AI

Detected by the catalog sweep, confirmed on the next sweep 61s later — 17 alerts delivered.

Elorian

Verified live 2026-10-16. Missed. Elorian was a new lab, and its entry was added already live, so it had nothing to fire. Elorian's docs listed elorian-1 on Oct 16; OpenRouter listed it at 03:10 UTC on Oct 17. Since Oct 24 the radar reads the labs' own model lists without a key (a public mirror of them, plus the labs' own docs and pricing pages) and emails the operator within a minute about any id there that no hunt covers, with a one-tap way to send the alert.

Meta

Verified live 2026-10-15. Missed. The Muse new-line hunt skipped any id containing "spark", and the Spark hunts only took version numbers, so the open-weights release matched nothing. Since Oct 24 the radar reads the labs' own model lists without a key (a public mirror of them, plus the labs' own docs and pricing pages) and emails the operator within a minute about any id there that no hunt covers, with a one-tap way to send the alert.

NVIDIA

Verified live 2026-10-15 — Oct 15 on Hugging Face (NVIDIA's model card) · 120B total / 12B active open MoE, text, image and video in, up to 1M tokens of context, OpenMDW-1.1 · our own sweep read it on four catalogs in 21 hours: nvidia/nemotron-3.5-super-vl on OpenRouter at $0.15/$0.50 per 1M, wandb at $0.25/$1.50, and models.dev's nvidia and opencode listings free. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

NVIDIA

Verified live 2026-10-15 — Bolt 74B-A9B on Hugging Face Oct 15, Bolt E4B (8B, 4B effective) on Hugging Face and build.nvidia.com Oct 23, per NVIDIA's model cards · 74B total / 9B active open MoE, up to 256K tokens of context, with a native 2-bit build that fits in 24GB, OpenMDW-1.1 · no bolt id has reached a catalog we sweep yet. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

xAI

Verified live 2026-10-14. Missed. There was no hunt for xAI's voice models, and they are not in the token-priced catalogs the radar swept, so nothing fired. xAI's own model docs, swept since Oct 24, list grok-voice-3.0. Since Oct 24 the radar reads the labs' own model lists without a key (a public mirror of them, plus the labs' own docs and pricing pages) and emails the operator within a minute about any id there that no hunt covers, with a one-tap way to send the alert.

Google

Detected by the catalog sweep, confirmed on the next sweep 720s later — 347 alerts delivered.

Meta

Verified live 2026-10-12. Missed. Meta's model list carried muse-wonder-1.0 from 16:27 UTC on Oct 12, but it was a new Muse line and no entry matched it, so no alert fired. Since Oct 24 the radar reads the labs' own model lists without a key (a public mirror of them, plus the labs' own docs and pricing pages) and emails the operator within a minute about any id there that no hunt covers, with a one-tap way to send the alert.

Anthropic

Detected by the catalog sweep, confirmed on the next sweep 61s later — 16 alerts delivered.

Mistral

Detected by the catalog sweep, confirmed on the next sweep 61s later — 17 alerts delivered. Late: a public mirror of Mistral's own model list had mistral-large-4-0 at 12:48 UTC, about 50 minutes before this alert, which fired once OpenRouter listed it. That mirror and Mistral's model docs are swept since Oct 24.

Google

Verified live 2026-10-06. Missed. Google listed gemini-nano-banana-2.1 at 15:00 UTC on Oct 6 and no entry covered the Nano Banana line yet, so no alert could fire. Since Oct 24 the radar reads the labs' own model lists without a key (a public mirror of them, plus the labs' own docs and pricing pages) and emails the operator within a minute about any id there that no hunt covers, with a one-tap way to send the alert.

Black Forest Labs

Detected by the catalog sweep, confirmed on the next sweep 60s later — 16 alerts delivered.

Ant Group

Detected by the catalog sweep, confirmed on the next sweep 60s later — 17 alerts delivered.

OpenAI

Detected by the catalog sweep, confirmed on the next sweep 60s later — 318 alerts delivered. Late: a public mirror of OpenAI's own model list had gpt-6.1-sol at 17:12 UTC, about 13 minutes before this alert. Our OpenAI source needs a key we don't have; OpenAI's model docs are swept since that day.

Anthropic

Detected by the catalog sweep, confirmed on the next sweep 60s later — 315 alerts delivered.

Meituan

Verified live 2026-09-27 — our sweep read nano-gpt/longcat-2.5-preview at $0.75/$2.95 per 1M, 1M ctx, on Sep 27, and a free relist of the same preview on Sep 26 · this row stands for the whole LongCat line, including the 2.0 our ledger has carried since July · that it is Meituan's, that LongCat-2.0 is a 1.6T MIT-licensed MoE with 48B active, and its benchmark figures, are search-grade and vendor-reported, not our own reading · the price we read is a gateway's, not necessarily Meituan's. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Perceptron

Verified live 2026-09-25 — our sweep read perceptron/perceptron-mk1.5 at $0.15/$1.50 per 1M, 36,864 ctx, on Sep 25, a first-party vendor slug rather than a gateway path · this row stands for the whole Perceptron line, including the Mk1 our ledger has carried unmapped · that it is Perceptron Inc's, that it is an embodied-reasoning model for robots, drones and glasses answering with points, boxes and polygons, and that it cuts Mk1's latency 2-5x, are search-grade, not our own reading · the weights licence we do not claim either way. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Fireworks AI

Verified live 2026-09-24 — our sweep read fireworks-ai/.../ember-1 at $3/$15 per 1M on Sep 24, the same sticker Fireworks launched it at that day · reasoning model post-trained from Kimi K3, ~40% fewer reasoning tokens at comparable quality on Fireworks' own evals · 1M ctx · announced as a research preview with a two-week serverless window, so it may not persist. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Anthropic

Detected by the catalog sweep, confirmed on the next sweep 61s later — 308 alerts delivered.

OpenAI

Verified live 2026-09-22. Missed. The radar sent no alert for this launch: GPT-6 Sol's catalog ids matched GPT-6 Astra's pattern, so they were filed under Astra, a live model, and never surfaced as new. Found Sep 24, two days late. Sol got its own entry that day, and live entries can no longer absorb a successor's ids.

OpenAI

Verified live 2026-09-22. Missed. The radar sent no alert for this launch: GPT-6 Luna's catalog ids matched GPT-6 Astra's pattern, so they were filed under Astra, a live model, and never surfaced as new. Found Sep 24, two days late. Luna got its own entry that day, and live entries can no longer absorb a successor's ids.

xAI

Detected by the catalog sweep, confirmed on the next sweep 60s later — 307 alerts delivered.

Xiaomi

Verified live 2026-09-21 — our sweep read xiaomi/mimo-v2.6-pro at $0.43/$0.87 per 1M and mimo-v2.6-flash at $0.14/$0.28, both 1.05M ctx, on Sep 21; the Pro-UltraSpeed tier at $4.35/$8.70 the same hour, and a free Flash listing at 200k ctx the next morning · that it is the open-source MiMo-V2.6 release (Pro ~1.02T / Flash 309B, MIT, 7,000+ RL environments) announced Sep 22, and that Artificial Analysis scores Pro at 46 on its Intelligence Index, is search-grade, not our own reading. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

StepFun

Verified live 2026-09-20. Missed. The sweep read stepfun/step-5-preview through two gateways the day it landed and it matched nothing: the armed entry was Step 4, a version StepFun skipped. Step 6 is now armed on StepFun's release cadence.

PrismML

Verified live 2026-09-18 — our sweep read prism-ml/ternary-bonsai-2-27b at $0.07/$0.50 per 1M, 262k ctx, on Sep 18 · a 1.58-bit ternary compression of Qwen3.8 27B rather than a model trained from scratch · that it is PrismML's, Apache 2.0, 5.9 GB and holds ~98% of the original is search-grade, not our own reading. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Circuit & Chisel

Verified live 2026-09-17 — our sweep read unbiased/pareto at $2.50/$7.50 per 1M, 262k ctx, on Sep 17 · a composite that routes across frontier and open models rather than a base model of its own · shipped first as the stealth id stealth/union-alpha, which our sweep read FREE at the same 262k ctx on Sep 16 · that it is Circuit & Chisel's, sold through their Unbiased platform, and that the stealth run was the same model, is search-grade, not our own reading. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Sakana AI

Verified live 2026-09-11 — an orchestration tier, not a base model: a router LM over a swappable pool of other models · our sweep read fugu-max at $2/$6 and fugu-ultra-v2 at $5/$30 per 1M, both served at 1M ctx, on Sep 11 · the Sep 11 launch date and the benchmark wins over Opus 5 and Fable 5 are Sakana's and the trade press's, not our own reading. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Shanghai AI Lab

Verified live 2026-09-11 — live in preview — agentic research model · our sweep read atria-dawn-preview listed FREE at 262k ctx on Sep 14; the ~744B MoE scale, the MIT weights on Hugging Face and the Sep 11 date are the trade press's and the model cards', not our own reading · the GA is still ahead. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

OpenAI

Verified live 2026-09-10 — voice · full-duplex speech-to-speech, twelve voices · in the API since Sep 10 at $0.05/min for the voice layer, backend model billed separately · date and price are from reporting, not from our own sweep: no gpt-live id is in the catalogs we read. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

DeepSeek

Verified live 2026-09-10 — Sep 10 · 1M ctx · $0.15/$0.60 per 1M as our sweep read it, which is DeepSeek's off-peak rate (peak hours on weekdays cost double, $0.30/$1.20, per its pricing page) · DeepSeek serves it under the unversioned id deepseek-flash, so the id itself never carries the version · Oct 17: upgraded in place to the 1017 checkpoint (DeepSeek-V4.1-Flash-1017, per DeepSeek's changelog), same deepseek-flash id and price, web and app updated too; open weights on Hugging Face as deepseek-ai/DeepSeek-V4.1-Flash-1017 (MIT, same config.json as the Sep 10 weights); OpenRouter lists it as deepseek/deepseek-v4.1-flash-1017 at $0.15/$0.60, 1M ctx, 384K max output · Oct 19: the Vals Index (updated Oct 18) scores the 1017 build 56.58%, #14 of 50 and the top open-weight model there, up from 51.32% for the Sep 10 build; $0.64 per test on Vals' run, against $0.33 for the Sep 10 build · Oct 19: DHS ordered DeepSeek products, and third-party products built on DeepSeek, excluded from and removed from US federal civilian agencies (Federal Register 2026-21447, publishing Oct 20) · Oct 20: DeepSeek open-sourced DeepInfer (Apache-2.0), the inference engine it serves V4.1 Flash with, for NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs · Oct 20: Vals says the 1017 build completed 16.3 of 21 objectives on its coming offensive-cybersecurity benchmark, against 15.7 for GPT-6 Sol, on several times the tokens (a first finding; the full evaluation is due in the coming weeks). No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Inception

Detected by the catalog sweep, confirmed on the next sweep 5772s later — 14 alerts delivered.

Sapiens AI

Verified live 2026-09-09 — our sweep read nano-gpt/agnes-3.0-flash at $0.05/$0.15 per 1M, 524k ctx, on Sep 9 · that the line is Sapiens AI's is search-grade, not our own reading · context window left unclaimed: trackers give 1M for the API model and 262k for a separate open-weight preview, and we read a third number. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

NEX AGI

Verified live 2026-09-08 — agentic family (mini · Pro · Max) · our sweep read the Pro and mini ids at 262k ctx on a free tier, so the standing paid rate is unknown and stays unquoted. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

OpenAI

Verified live 2026-09-08 — image · Flare and Sunburst, the two API tiers of ChatGPT Images 2.5 · both listed at $5/$30 per 1M on the catalog we read. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

OpenAI

Detected by the catalog sweep, confirmed on the next sweep 69s later — 110 alerts delivered.

IFM

Verified live 2026-09-03 — fully open fleet (weights, code and training data) at 375B-A23B · 36B-A4B · 32B · 7B · 3.7B · 0.9B; the launch date is from IFM's own Sep 3 announcement, not our own reading · our sweep read the 7B at $0.05/$0.20 per 1M served at 131k ctx, against a 524k native window IFM claims. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Meta

Detected by the catalog sweep, confirmed on the next sweep 61s later — 14 alerts delivered.

Google

Verified live 2026-09-02. Missed. The radar read gemini-3.8-flash at 09:31:44 UTC on Sep 2 but classed it as a variant of the next-Flash entry, so the primary never fired and no alert went out. The classifier now judges each id against the entry it is checking (fixed that day).

Anthropic

Detected by the catalog sweep, confirmed on the next sweep 60s later — 110 alerts delivered.

Anthropic

Verified live 2026-09-01 — live - shipped Sep 1 with Fable 5.1 · verified organizations only: life-science teams through the Life Sciences Verification Program (Sep 17), security teams through the three-tier Cyber Verification Program (Oct 6). No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Tencent

Verified live 2026-08-28 — live in preview — 770B total / 49B active MoE · Apache 2.0 open weights · $0.83/$2.50 per 1M · 1M ctx · the GA flagship is still ahead. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Google

Verified live 2026-08-26 — $2/$12 per 1M · speech-to-text, 85+ languages · a streaming `-live` id is listed alongside it. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Alibaba

Verified live 2026-08-26. Missed. No alert fired: until Sep 2 the radar classed every Flash-family id as a variant of its own entry, so no Flash launch ever fired as primary. This one was verified by hand; the classifier was fixed Sep 2.

Zhipu AI

Verified live 2026-08-26. Missed. No alert fired: until Sep 2 the radar classed every Flash-family id as a variant of its own entry, so no Flash launch ever fired as primary. This one was verified by hand; the classifier was fixed Sep 2.

IBM

Verified live 2026-08-25 — Apache 2.0 open weights, 3B/8B/30B · switchable reasoning · 8B listed at $0.10/$0.15 per 1M, 131k ctx. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Alibaba

Verified live 2026-08-24 — video · 30-second single-pass clips from text, image, audio, video or documents · hosted on Alibaba Cloud. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

DeepReinforce

Verified live 2026-08-19 — 397B / 35B MoE + 9B dense · MIT · open weights on Hugging Face. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Zhipu AI

Verified live 2026-08-14. Missed. The GLM-5.3 entry was armed and its pattern was right, but the model launched on Z.ai's own API and no catalog we swept carried the id, so it never fired. models.dev became a launch source that day.

Google

Verified live 2026-08-13. Missed. Google listed gemini-3.7-flash at 17:03 UTC on Aug 13 and no registry entry matched it, so no alert could fire; we learned of it from outside the radar. Since then the next Flash version is armed before it ships.

xAI

Detected by the catalog sweep, confirmed on the next sweep 63s later — 106 alerts delivered.

Motif Technologies

Verified live 2026-08-12 — our sweep read aiand/motif-technologies/motif-3 at $0.5/$2 per 1M, 262k ctx, on Aug 12 · that Motif Technologies is a Korean lab, that this is a long-context MoE with extended thinking, and that the weights are on Hugging Face is search-grade, not our own reading. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

NVIDIA

Verified live 2026-08-11 — Aug 11 on Hugging Face (NVIDIA's model card) · 30B total / 3B active open MoE, up to 1M tokens of context · our own sweep read it six times across five catalogs on Aug 11-12: models.dev's nvidia, wandb, opencode and nano-gpt listings plus OpenRouter's nemotron-3.5-lightning:free · the pattern also takes opencode's misspelled nemotron-3.5-lighting-free. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Meta

Verified live 2026-08-09 — $0.35/$1.50 per 1M · the small Muse. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Upstage

Verified live 2026-08-07 — our sweep read upstage/solar-pro4 at $0.3/$1.2 per 1M, 524k ctx, on Aug 7, and upstage/solar-mini4 at $0.05/$0.20, same 524k ctx, on Sep 23 · this row stands for the whole Solar line · that Upstage is a Korean lab and that Mini 4 is a 35B MoE with 3B active is search-grade, not our own reading. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

ByteDance

Verified live 2026-08-07 — video · public API since Aug 7, after a Jun 23 announcement (reported) · up to 30-second single-pass clips, up to 50 references, up to 4K (reported) · OpenRouter lists bytedance/seedance-2.5 with no per-second price. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Meta

Verified live 2026-08-05. Missed. Muse Spark 1.2 matched no entry: 1.1 was live and Muse Spark 2 was armed, with nothing for the 1.x point releases in between. Found Aug 5 through Meta's Muse Code news; 1.3 to 1.9 were armed that day, and 1.3 fired on Sep 2.

Meta

Verified live 2026-08-05 — live (beta) — Meta's first terminal coding agent, via Meta Model API. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Alibaba

Detected by the catalog sweep, confirmed on the next sweep 61s later — 103 alerts delivered.

Thinking Machines

Verified live 2026-07-30 — 276B total / 12B active · SWE-bench Verified 80.2% · 512K ctx · open weights. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

MiniMax

Verified live 2026-07-30 — video · #1 video editing, #2 text-to-video · open weights · API price TBA. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

xAI

Verified live 2026-07-29 — speech-to-speech · 82.9% on the Artificial Analysis index, up from 75.7% · grok-voice-latest repointed to it Aug 5 · deprecated Oct 14, replaced by Grok Voice 3.0 (xAI release notes). No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Anthropic

Detected by the catalog sweep, confirmed on the next sweep 61s later — 95 alerts delivered.

Black Forest Labs

Verified live 2026-07-23 — Video went GA Aug 4 via BFL API · 20s at 1080p, native audio, lip-sync in 14+ languages · Dev open weights still unreleased. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Ant Group

Verified live 2026-07-23 — 124B MoE · 262k ctx · free on the router through Aug 3, then priced. No launch detection is logged under this model's id. If a hunt under another name caught it, that hunt's row says so; otherwise the radar sent no alert for it. We don't reconstruct timestamps.

Google

Verified live 2026-07-21 — $0.30/$2.50 per 1M; high-throughput agentic search + doc processing. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Google

Verified live 2026-07-21 — $1.50/$7.50 per 1M, cheaper than 3.5 Flash; 17% fewer output tokens. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Moonshot

Verified live 2026-07-15 — dropped Jul 15 · 2.8T · open weights landed Jul 27 (1.56 TB, MXFP4). Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Thinking Machines

Verified live 2026-07-15 — open weights · 975B MoE · dropped Jul 15. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

OpenAI

Verified live 2026-07-09 — Sol · Terra · Luna since Jul 9 · repriced Jul 30: Luna −80%, Terra −20%. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

xAI

Verified live 2026-07-09 — public Jul 9. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Meta

Verified live 2026-07-09 — Meta Model API, Jul 9. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Meta

Verified live 2026-07-07 — Meta Superintelligence Labs' first image model, shipped Jul 7; listed on the API catalogs Aug 26. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Anthropic

Verified live 2026-07-01 — returned Jul 1. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Anthropic

Verified live 2026-06-30 — shipped Jun 30. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Google

Verified live 2026-06-30 — video generation and conversational editing · preview Jun 30, GA Aug 27 as gemini-omni-1.1-flash · $1.50 per 1M input, $17.50 per 1M video output, about $0.10 a second of 720p (Google's pricing page, read Oct 9). Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Zhipu AI

Verified live 2026-06-16 — $0.40/$2.52 per 1M · no longer the current shipped GLM: 5.3 landed Aug 14 and 5.3 Flash Aug 26. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Google

Verified live 2026-06-10 — our sweep read google/diffusiongemma at $0.05/$0.15 per 1M, 262k ctx, on Sep 19 · that it is Google DeepMind's first open-weight text-diffusion model (26B-A4B MoE on the Gemma 4 architecture, Apache 2.0, ~4x faster by denoising 256-token blocks in parallel) and that it was announced Jun 10 is search-grade, not our own reading. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Anthropic

Verified live 2026-06-09 — shipped Jun 9 with Fable 5, export-controlled Jun 12, controls withdrawn Jun 30; on Azure at $10/$50, not on Bedrock · Oct 21: Dario Amodei on X says Anthropic first had Mythos-class models in February and held general availability until June, when stronger cyber safeguards were ready, offering them only to a limited set of partners for cyberdefense until then · Oct 23: Anthropic's analysis of the AISI cyber-test incident says Mythos 5 started the GitHub attack in 83% of replays, against 0-13% for Mythos 5.1, Opus 5.5 and an unnamed internal research model (Anthropic's own replays). Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Cohere

Verified live 2026-06-09 — live - Cohere's second product line, a coding model - listed at $0/$0 per 1M, 256k ctx / 64k out - reasoning - open weights. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

MiniMax

Verified live 2026-05-31 — $0.30/$1.20 per 1M. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Anthropic

Verified live 2026-05-27 — live. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Cohere

Verified live 2026-05-20 — live - $2.50/$10 per 1M, 128k ctx / 64k out - reasoning - open weights. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Google

Verified live 2026-05-19 — preview since I/O, on OpenRouter · deprecated by Google Oct 8: no longer selectable in AI Studio, API requests to gemini-3.5-flash auto-route to gemini-3.6-flash, and the endpoint is not shutting down (Gemini API changelog + deprecations page, read Oct 26). Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Ant Group

Verified live 2026-05-08 — 1T reasoning line, $0.07/$0.63 per 1M. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

OpenAI

Verified live 2026-04-24 — live. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

DeepSeek

Verified live 2026-04-24 — live · open weights · superseded by V4.1 Flash — per DeepSeek's own changelog rather than our own reading, deepseek-v4-flash routes to V4.1 Flash; the matching deepseek-v4-pro cutover set for 04:00 UTC Sep 14 was reversed on its eve and V4 Pro is still served at unchanged billing, which is search-grade too — but our own bench refresh still scored deepseek-v4-pro-0813 as a separate model on Sep 16 · our sweep read the bare deepseek-flash alias on Sep 10 at $0.15/$0.60 per 1M, 1M ctx. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Xiaomi

Verified live 2026-04-22 — live - $0.14/$0.28 per 1M, cache reads $0.0028 - 1.05M ctx / 131k out - text, image, audio and video in - open weights - Pro and Pro-UltraSpeed tiers followed, the latter Jun 8. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Google

Verified live 2026-04-02. Missed. Gemma 4 shipped Apr 2 and its entry sat on 'expected' for four months: the pattern rejected Google's real ids (gemma-4-31b-it), so the armed entry could never match. Found Aug 3 by the armed-entry audit, which now runs every armed pattern against every id the catalogs list.

xAI

Verified live 2026-03-10 — live since March — multi-agent, reasoning and non-reasoning SKUs; 4.20-beta variants are still under test on the arena board. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Inception

Verified live 2026-02-24 — diffusion LLM · listed in the catalogs we sweep; launch date is from Inception's own Feb 24 announcement, not our own reading. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Amazon

Correction: this was not a launch. Nova 2 Pro was announced at re:Invent on Dec 2, 2025 and is still in preview for Nova Forge customers. What the sweep saw on Sep 15 was a gateway listing the clean id nova-2-pro-v1 for the first time — Amazon's own catalog dates that model to 2025-12-03. The alert should not have fired, and 14 subscribers received it. The guard that now holds a detection the catalog dates as old was built from this.

Runway

Verified live 2025-12-01 — text- and image-to-video · announced Dec 1 2025, on all Runway paid plans Dec 11 · $0.12 a second on Runway's API docs (read Oct 9) · listed on OpenRouter as runway/gen-4.5. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.

Google

Verified live 2025-10-15 — video with native audio · public preview on the Gemini API Oct 15 2025 with Veo 3.1 Fast; Lite from Mar 31 2026 · per second: $0.40 (720p/1080p) or $0.60 (4K); Fast $0.10-$0.30; Lite $0.05-$0.08 (Google's pricing page, read Oct 9) · Oct 22 2026: the three Gemini API PREVIEW ids shut down TODAY — veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview — and Google names gemini-omni-1.1-flash the replacement, on a different API (Interactions, not predictLongRunning), so the three per-second prices above are Gemini API preview prices for ids that no longer answer. SEARCH-GRADE: ai.google.dev is egress-blocked from the sweep sandbox and the deprecations page was NOT opened; four independent repos migrating off it and Google's own Veo docs are cited for the Oct 22 date. Row stays LIVE because Vertex AI is unaffected — veo-3.1-generate-001 is GA there with no announced retirement (reported as 'Nov 17 2026 or later') — and because the model is still served. Our own sweep has never read a veo id in any catalog: no wire item, nothing in seenIds, so every Veo fact on this row is attribution, not our reading. NOT RE-PRICED: no source gives a Vertex per-second price we could substitute, and whether Lite survives anywhere is unconfirmed, so the Lite figure on the board card is flagged for Mark rather than guessed at. Detection timestamps for this launch are offline: the one-shot ledger was reset Jul 21, 2026 during a store migration and re-imports from the archived database in early August. Until then this row shows only the verified live date — we don't reconstruct timestamps.