Timeline

google/gemma-4-31b-it:free is listed in the Kilo gateway catalogue at $0/$0 with no announced end (expiration_date null); 262K context via google-ai-studio, may train on prompts

thinkingmachines/inkling:free is listed in the Kilo gateway catalogue at $0/$0 until 2026-11-04 (gateway expiration_date); 1M context, may train on prompts → end date Nov 4

opencode/pulse-alpha-free is listed in the Zen catalogue at $0 (stealth model, free for a limited time to collect feedback, no end date published); 1M context; EU and other regions excluded

poolside/laguna-s-2.1:free and poolside/laguna-xs-2.1:free are no longer in the OpenRouter catalogue (452 models, zero laguna on re-read 2026-11-01, same snapshot hash) nor in the Kilo gateway catalogue (384 models, zero laguna on re-read 2026-11-01); no end date announced, true removal window 2026-10-27~11-01 for the s-2.1 arms and 2026-10-31~11-01 for the xs-2.1 arms → end date Nov 1

poolside/laguna-s-2.1-free is no longer in the Vercel AI Gateway catalogue at the 2026-10-31T09:00Z daily check, and still absent when we re-read the same API the same day; only the paid poolside/laguna-s-2.1 remains; no end date announced, true removal window 2026-10-27~31 → end date Oct 31

Ling 3.1 Flash 1019 listed on two free arms: Vercel AI Gateway carries inclusionai/ling-3.1-flash-1019 at $0/$0 (changelog: free until November 3, 2026) and Kilo Code carries inclusionai/ling-3.1-flash-1019-free at $0/$0 (Novita AI, free until November 3, 2026; may train on prompts). New HF weights repo (created 2026-10-27) coexists with the old id, so a separate model entry. → end date Nov 3

opencode/jev-1.13-free is no longer in the Zen catalogue (78 models, zero Jeeves entries on re-read 2026-10-30, down from 82 on 2026-10-29) and absent from the models.dev opencode provider list (130 models, zero Jeeves); no end date announced, true removal window 2026-10-23~30 → end date Oct 30

SiliconFlow rendered pricing table lists zai-org/GLM-5.3-Flash at ¥0.80/¥2.80 per M tokens (cache ¥0.23), contradicting the model plaza price field "0"; treated as paid, not free

Added hunyuan-mt-7b-siliconflow: translation-specialised chat model, rendered pricing table marks it 免费 (strong evidence); translation models count as text LLM lanes

Added hy3-codebuddy: Hy3 is 限时免费 all day inside the official CodeBuddy / WorkBuddy clients (CN site, one shared credit pool). Own check the same day: the CLI model table served the multiplier x0.00 and 5 one-shot runs (~120k tokens) took 0 credits. The official mini-program model table has labelled Hy3 限时免费 since its 2026-07-22 revision

Added hy4-preview-codebuddy as listed: only the night window 23:00-08:00 (Beijing) carries a free quota, so the window itself is untested here. Daytime read the same day: the CLI model table served Hy4 preview at x0.29 and the model answered. The official notice adds that Hy4 preview is a text-only model, so an image or video sub-task it hands over to another model is billed normally

Buddy AI 国际站 official billing page (last updated 2026-08-07, read 2026-10-28) fixes how the Free grant is claimed: daily active bonus Free 30 / Pro 50 credits, at most once per activity day, credited the next day, and it stacks with the monthly credits; a day counts only if a chat or task is started in the WorkBuddy or CodeBuddy client (web chats do not count); scripted, emulated or bulk-registered activity is excluded; Free's 100/month is not issued while Pro or the 7-day Pro trial is active; the activity's end date is announced separately

opencode-go/iceman-contributor-free answered HTTP 400 Model is unavailable on 9 probes across 4 rounds (2026-10-27 15:3x - 2026-10-28 15:3x, seats go2/go3); the Go catalogue still lists it, so unavailable per ADR-022 section 4, not ended

Freebuff's free plan no longer quotes a money allowance: at the 2026-10-27 recheck the pricing page lists Free $0/month with "30 hours refilled daily" and says "Usage is measured in hours", with a per-model multiplier and daily hours resetting at midnight in the local timezone; the 2026-10-08 record read "$1.50 refilled daily" measured in Freebucks, and no hours-to-money rate is published

TRAE CN's free-user sign-up grant is 2,000 credits at the 2026-10-27 recheck (新用户注册福利 一次性赠送 2000 积分, valid 31 natural days) where our 2026-10-08 record read 4,000; on the same page the daily check-in is still 100 credits a day and the monthly login gift still 500, and the 4,000 figure now appears only as the paid 会员 Pro monthly allowance

Google's Antigravity plans page now describes the free baseline as "Use of Gemini models, including Gemini 3.8 Flash and other offered Gemini Enterprise models as the core agent model" plus unlimited Tab completions and all product features; Gemini 3.1 Pro, named in our 2026-10-08 record, is no longer named there - the page refers model availability to its Models list, so our record stops naming it rather than claiming it was removed

nvidia/nemotron-3-super-120b-a12b:free is no longer in OpenRouter's free catalogue at the 2026-10-27T13:33Z daily check, and still absent when we re-read the same API the same day; only the paid nvidia/nemotron-3-super-120b-a12b remains, at $0.30/$2.40 per Mtoken; no end date announced → end date Oct 27

Zhipu's official news (zhipuai.cn/zh/news/148, 2026-01-19) states GLM-4.7-Flash replaces GLM-4.5-Flash free and GLM-4.5-Flash retires 2026-01-30 with requests auto-routed to GLM-4.7-Flash; BigModel's free-model price table now lists GLM-4.7-Flash and no longer 4.5-Flash; lane ended at 2026-01-30 → end date Jan 30

Tencent removed hunyuan-lite (Hunyuan-lite) from the Hunyuan free-quota table and price list (cloud.tencent.com/document/product/1729/97731, re-checked 2026-10-27; the free-quota table is client-rendered, the removal is from the sweep read). Freetokens tracks Hunyuan free as the first-time 1M-token resource package, which is unaffected; there is no hunyuan-lite lane to mark ended.

iceman-contributor-free is in the OpenCode Go model list: $0 on input, output and cache reads and "free for a limited time" per the model registry, but it sits behind the $10/mo Go subscription and does not draw down the plan's quota, carries no maker (OpenCode's own data page lists the provider as "Unknown"), has no row in the Go pricing or privacy tables so its training terms are unstated, and has no Artificial Analysis score

minimax/minimax-m3.1-flash:free is no longer in OpenRouter's free catalogue at the 2026-10-23T08:00Z daily check, and still absent when we re-read the same API the same day; the paid minimax/minimax-m3.1-flash remains at $0.15/$0.60 per Mtoken; it was added 2026-10-20 and was still listed on 2026-10-20; we never called it; OpenRouter's model page states "The free launch period ended Oct 22, 2026" → end date Oct 22

opencode/minimax-m3.1-flash-free is no longer in the OpenCode Zen free model list at the 2026-10-23T08:00Z daily check, and still absent when we re-read the same list the same day; it was added 2026-10-20 and was still listed on 2026-10-20; we never called it; no end date announced → end date Oct 23

opencode-go/minimax-m3.1-flash-free is no longer in the OpenCode Go free model list at the 2026-10-23T08:00Z daily check, and still absent when we re-read the same list the same day; the list keeps the bare minimax-m3.1-flash paid entry; it was added 2026-10-20 and was still listed on 2026-10-20; we never called it; no end date announced → end date Oct 23

minimax/minimax-m3.1-flash is no longer free on the Vercel AI Gateway at the 2026-10-23T08:00Z daily check: the models API still lists the ID but prices it at $0.15/$0.60 per Mtoken instead of $0 (still absent when we re-read the same API the same day); it was listed at $0 on 2026-10-20; we never called it; no end date announced → end date Oct 23

thinkingmachines/inkling-small:free is no longer in Kilo's model catalogue at the 2026-10-23 catalogue recheck (the 401-model list keeps 13 other :free ids); the same API holds thinkingmachines/inkling-small as paid-only, isFree false at $0.45/$1.00 per Mtoken, and mastra.ai's Kilo provider mirror lists the same id at those prices with no :free variant; first seen in the catalogue 2026-08-21, last listed 2026-10-06; it never answered one of our calls; no end date announced → end date Oct 23

Google deprecated gemini-3.7-flash on 2026-10-08: it is no longer available to select in Google AI Studio and its requests auto-route to gemini-3.8-flash; at the 2026-10-23 catalogue recheck the id is absent from the free-tier endpoint table and from the pricing page, so the free tier is gone; second source is the 2026-10-08 entry in the AI Studio changelog; first seen 2026-08-14, last listed 2026-10-06 → end date Oct 8

Google deprecated gemini-3.5-flash on 2026-10-08 in favour of gemini-3.6-flash: it is no longer available to select in Google AI Studio; at the 2026-10-23 catalogue recheck the id is absent from the free-tier endpoint table and from the pricing page, so the free tier is gone; second source is the 2026-10-08 entry in the AI Studio changelog; first seen in our records 2026-10-06, last listed the same day → end date Oct 8

Mistral-7B-Instruct-v0.3 is no longer in OVHcloud's keyless model catalogue at the 2026-10-23 recheck (the 22-model anonymous-lane list does not name it), and OVH's public AI Endpoints catalogue page carries no entry for it either; it was last listed 2026-10-08; no end date announced → end date Oct 23

OVH has put Mistral-Small-3.2-24B-Instruct-2506 on its end-of-life schedule: at the 2026-10-23 recheck the REST catalogue has available false with model_eol_date 04/11/2026, and the public catalogue page badges it Being retired with end of life on 4 Nov 2026 (OVH's own wording is that a retired model keeps answering until its EOL); the anonymous 2-requests-a-minute lane is still published for it, so the lane stays listed with an official end date; recheck at 2026-11-04 → end date Nov 4

OVH has put Mistral-Nemo-Instruct-2407 on its end-of-life schedule: at the 2026-10-23 recheck the REST catalogue has available false with model_eol_date 04/11/2026, and the public catalogue page badges it Being retired with end of life on 4 Nov 2026 (OVH's own wording is that a retired model keeps answering until its EOL); the anonymous 2-requests-a-minute lane is still published for it, so the lane stays listed with an official end date; recheck at 2026-11-04 → end date Nov 4

DeepSeek V4 Flash on Letta is no longer free: at the 2026-10-23 recheck the model catalogue lists it under the credit-billed tier, and a live free-plan probe was refused with "minimum $1 in credits is required" (the same error as its sister lane on 2026-10-17); a letta/echo control on the same account passed, so the refusal is model-specific → end date Oct 23

Flint is on OpenRouter's free catalogue as recursive/flint-v1.0:free ($0 per token on the models API, 262,144 context, text input only): first seen by the daily free-catalogue check at 2026-10-22T08:00Z and still there when we re-read the same API the same day; we have not called it

Flint is on OpenCode Zen's free model list as opencode/flint-40b-free: first seen by the daily free-catalogue check at 2026-10-22T08:00Z and still there when we re-read the same list the same day; we have not called it

opencode/longcat-2.5-preview-free is no longer in the OpenCode Zen free model list at the 2026-10-22T08:00Z daily check, and still absent when we re-read the same list the same day; it was added 2026-09-26, answered 10 parallel runs on 2026-10-06 and was still listed on 2026-10-20; no end date announced → end date Oct 22

opencode-go/longcat-2.5-preview-free is no longer in the OpenCode Go free model list at the 2026-10-22T08:00Z daily check, and still absent when we re-read the same list the same day; it was added 2026-09-26, answered 15 parallel runs on 2026-10-04 and was still listed on 2026-10-13; no end date announced → end date Oct 22

BigModel docs banner says GLM-4.5-Flash retires 2026-01-30 (auto-route to GLM-4.7-Flash) but page still live and self-contradictory; Z.ai model doc 404, pricing still lists Free; inference untested (needs key); lane set to unavailable pending verification → end date Jan 30

opencode/space-bunny-free is no longer in the OpenCode Zen free model list at the 2026-10-20T08:00Z daily check; the ID was there on 2026-10-13; no end date was ever published, so the free window closed without notice → end date Oct 20

opencode/muse-spark-1.3-contributor-free is no longer in the OpenCode Zen free model list at the 2026-10-20T08:00Z daily check; it was there, and answering 10 parallel agent runs, on 2026-10-07; no end date announced → end date Oct 20

opencode/muse-spark-1.2-contributor-free is no longer in the OpenCode Zen free model list at the 2026-10-20T08:00Z daily check; last in the catalogue on 2026-10-06, when our API call could not reach it; no end date announced → end date Oct 20

opencode/cat-jam-free is no longer in the OpenCode Zen free model list at the 2026-10-20T08:00Z daily check; it was added 2026-10-13 and answered 5 parallel runs the same day; no end date announced → end date Oct 20

opencode-go/cat-jam-free is no longer in the OpenCode Go free model list at the 2026-10-20T08:00Z daily check; it was listed on 2026-10-13 and never reached by our own calls; no end date announced → end date Oct 20

opencode/ll-alpha-free is no longer in the OpenCode Zen free model list at the 2026-10-20T08:00Z daily check; first seen 2026-10-16, never load-tested by us; no end date announced → end date Oct 20

stealth/cat-jam is no longer on the Vercel AI Gateway free list at the 2026-10-20T08:00Z daily check: the models API either drops the ID or stops reporting $0 input and output; it was listed at $0 on 2026-10-15 and 2026-10-16; no end date announced → end date Oct 20

MiniMax M3.1 Flash is on OpenRouter's free catalogue as minimax/minimax-m3.1-flash:free ($0 per token on the models API): first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

MiniMax M3.1 Flash is on OpenCode Zen's free model list as opencode/minimax-m3.1-flash-free: first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

MiniMax M3.1 Flash is on OpenCode Go's free model list as opencode-go/minimax-m3.1-flash-free: first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

MiniMax M3.1 Flash is on the Vercel AI Gateway free list as minimax/minimax-m3.1-flash at $0 input and $0 output: first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

Beam 1.0 is on OpenCode Zen's free model list as opencode/beam-free: first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

Beam 1.0 is on OpenCode Go's free model list as opencode-go/beam-free: first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

Beam 1.0 is on the Vercel AI Gateway free list as reflection/beam-free, shown as "Beam 1.0 (Free)": first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

Xing 4.0 29B is on OpenRouter's free catalogue as xingchen-agi/xing4.0-29b-a4b:free ($0 per token on the models API): first seen by the daily free-catalogue check at 2026-10-20T08:00Z; we have not called it

Still in Kilo's $0 model catalogue, but 3 anonymous calls failed 11:43-11:48Z with an upstream 404 from the Nvidia provider ("Function ... Not found for account")

inclusionai/ling-3.0-flash-sante:free is gone from Kilo's gateway catalogue and model site; 3 anonymous calls returned 404 "The requested model ... does not exist" (11:43-11:47Z) → end date Oct 20

stepfun/step-3.7-flash:free is gone from Kilo's gateway catalogue (the paid stepfun/step-3.7-flash remains); 3 anonymous calls returned 404 "The requested model ... does not exist" (11:44-11:47Z) → end date Oct 20

Letta's free window for DeepSeek V4.1 Flash ended 2026-10-17: model requests now consume credits, and a free account with a $0.00 balance was refused with "minimum $1 in credits is required"; the pricing docs say model requests consume credits at a rate depending on model API pricing, and the per-model price page needs a login → end date Oct 17

OpenRouter added deepseek/deepseek-v4.1-flash-1017 (created 2026-10-17T08:27:04Z): DeepSeek first-party endpoint only, no :free variant; DeepSeek upgraded V4.1 Flash in place to the 1017 snapshot with the API name (deepseek-flash) and pricing unchanged, so channels serving the unversioned deepseek-v4.1-flash id are on 1017 while their labels still say the old name

DeepSeek V4.1 Flash upgraded in place to 1017; the API name deepseek-flash and the price are unchanged. Which snapshot each third-party channel serves is not recorded.

Command Code delisted the free variant inclusionai/ling-3.1-flash:free; the model page now bills $0.30/$0.90/$0.06 per 1M (input/output/cache read, Go plan and above); no end date announced - the "free while it lasts" deal was replaced by a boosted credits deal (GOAT $100 / Pro $110) → end date Oct 16

GLM 5.3 Flash is free on OpenCode Zen: listed Free for input, output and cached reads in the Zen price table and present in the model catalogue; three coding-agent runs on 2026-10-16 cost nothing (two passed, the third aborted on an upstream rate limit)

DeepSeek V4.1 Flash is free on OpenCode Zen: listed Free for input, output and cached reads in the Zen price table, in the model catalogue, and three coding-agent runs on 2026-10-16 passed at zero cost

NVIDIA Nemotron 3.5 Super VL is free on OpenCode Zen: in the model catalogue and listed Free in the Zen price table, "available on OpenCode for a limited time"; the NVIDIA free endpoints log use for product improvement

LL Alpha is free on OpenCode Zen: a stealth model, "free on OpenCode for a limited time", listed Free in the price table and present in the model catalogue; collected data may be used to improve the model and is retained 30 days

Apodex 1.1 Mini is gone from OpenRouter: the model list holds no apodex ID any more and the endpoints API returns 0 endpoints for both the standard and the :free ID, so nothing serves it → end date Oct 16

OpenCode Zen dropped Step 5 Preview Free: not in the model catalogue, not in the Zen model list or price table, hidden from the client, and the free-tier endpoint answers "Model is unavailable" (HTTP 400) where live free models answer 403 FreeTierError → end date Oct 16

OpenCode Go dropped Step 5 Preview Free: the plan's model list keeps the paid Step 5 Preview ($1.00/$2.70) with no free variant, the free ID is out of the catalogue, and an authenticated call answers 400 "Upstream request failed: Model is unavailable." (0.37s) while a control free model still answered 200 → end date Oct 16

Ant Ling's billing upgrade notice sets the end of the daily 500K-token free allowance: 2026-10-31 23:59:59 (Beijing time), after which free credits become a ¥10 voucher issued on the 1st of each month (accounts that upgrade early get a one-off ¥20 voucher, valid 60 days); the official API still answered 200 on 2026-10-16 (usage 36/30/66 tokens, no cost field in the response) → end date Oct 31

Ant's own pricing page now bills Ling-3.1-flash pay-as-you-go at ¥0.50/¥1.50 per 1M with a limited-time 75% discount (list ¥2.00/¥6.00), replacing "限时免费 2 周"; the post-trial pricing is published after "rates to be announced" → end date Oct 14

OpenCode Zen dropped Ling 3.1 Flash: no longer in the free model catalogue or the paid one, and the chat endpoint answers "Model is unavailable" (a control free model still answered 403 FreeTierError); no changelog entry → end date Oct 15

Nemotron 3.5 Super VL (free) appears on OpenRouter at $0 prompt and $0 completion, served by an NVIDIA endpoint with 1M context and 131K max output (text, image and video input)

Cat Jam is now available in stealth for free on Vercel AI Gateway: $0 input and $0 output, 1M context, 131K max output, release date 10/15/2026; prompts and responses may be used for training

Ant's official two-week trial ended 2026-10-13: OpenRouter bills $0.075/$0.225 per 1M (75% off promo), Kilo Code and Vercel AI Gateway $0.30/$0.90; Vercel's -free model ID stops serving and OpenRouter's free variant is gone from the catalogue → end date Oct 13

Still $0.00/M on input, output and cache after the trial ended elsewhere: "free while it lasts", credits never draw down, available on the Go plan and above

Nemotron 3.5 Lightning Free joins Vercel AI Gateway at $0 input and $0 output, 1M context and 65K max output, release date 10/14/2026; the provider may retain prompts and responses for model improvement

Free :free variant removed from OpenRouter (pricing>0); paid version remains

opencode/fledge-alpha-free no longer in the Zen model list

opencode/nemotron-3-ultra-free no longer in the Zen model list

opencode/ling-3.0-flash-fin-free no longer in the Zen model list

cat-jam-free found in channel model list (watch)

cat-jam-free found in channel model list (watch)

Cline free promotion ended: API returns "Free model promotion ended"; models removed from Cline recommended models catalogue

Still listed in the Zen catalogue; a call returned HTTP 410 ModelDeprecated: Exo

First seen in the OpenCode Zen free catalogue; the listing may predate this check: Step 5 Preview

First seen in the OpenCode Go free catalogue; the listing may predate this check: Step 5 Preview

qwen/qwen3.8-27b:free is no longer in OpenRouter's model list; only the paid qwen/qwen3.8-27b remains

No longer in OpenCode Zen's model list (free catalogue check)

Not in Kilo's free model list any more: the signed-in Kilo CLI (7.8.3) lists 14 free models, none of these; most remain only as paid models

Not in Kilo's model list any more (signed-in Kilo CLI 7.8.3: model not found)

Early access on AI Gateway for Pro or Enterprise teams with purchased credits; token price $0 during the stealth period: Glyph Cluster

Retested through the OpenCode CLI without an account: 1.3 timed out after 180 s, 1.2 returned "Unexpected server error"; both stay listed. Possibly needs a signed-in account that grants the training permission (not verified)

Calls through OpenCode Go return "Unexpected server error" from 05:37 UTC; OpenCode Go's model list now has space-bunny without the -free suffix (a paid, seat-quota model), not space-bunny-free (checked 2026-10-06). Still free on OpenCode Zen

Both carry a green "Free" badge in the Letta app's Hosted model list (other hosted models show "Quota" or "Out of credits"); V4.1 Flash ran an agent session on a free account. No official start or end date found

Answered a test prompt through the Kilo CLI (7.8.3, roll-call) without signing in

Listed free in the Kilo CLI; the first test call hit "Daily limit reached … Credits don't affect this cap"

Answered a test prompt in the Cline CLI (3.0.68, signed in, plan mode)

In the Cline CLI's free-model list; the test call returned no text or an upstream error (429). Qwen 3.8 27B (404) and Step 3.7 Flash (not found) are listed but gone

Answered a test prompt through the OpenCode CLI (opencode run) with no account

In OpenCode Zen's free list (-free models); our test call failed (server error, upstream overload or timeout)

Kilo marks inclusionai/ling-3.1-flash free on purpose (Kilo-Org/cloud PR #7114); it answers without signing in and passed headless agent tasks, with an upstream 429 under load

On Freebuff's "Free hours per day, by model" table; not tested by us because Freebuff's terms forbid scripted use

Listed at $0 in Z.ai's own catalogue (standing free tier); not called by us yet

Listed at $0 in Zhipu BigModel's own catalogue (standing free tier); not called by us yet

Listed at $0 in Vercel AI Gateway's own catalogue (standing free tier); not called by us yet

Listed at $0 in SiliconFlow's own catalogue (standing free tier); not called by us yet

Free of charge on Google AI Studio's free tier (standing); not called by us yet

Tested with a Vercel AI Gateway key: Ling 3.1 Flash (Free) answers but is capped at 5 requests per minute per team (per region); Laguna S 2.1 Free returned 503

After buying AI Gateway credits (paid tier removes Vercel's per-model limit) the free Ling 3.1 Flash still costs $0 but the Novita upstream returned 503 for most calls: 10/10 once, then mostly 503 at 3–15 parallel

No longer in the Kilo CLI's model list (kilo models, 7.8.3); Kilo had announced Oct 5 as the end

Conditions found: the Contributor models need a signed-in Zen account with an explicit data-training opt-in, use the Responses API, and are limited to Meta's allowed regions (many RegionError reports outside the US); this explains our anonymous test failures

Listed at $0 on AIHubMix

Answered through the Qoder CLI (qodercli 1.1.65, headless -p) on a free personal account: 16 of 16 multi-step agent tasks passed at up to 10 parallel, every run 0 credits

Tested with a new NVIDIA build key: the key works (Nemotron 3 Super answered in 0.6 s) but GLM 5.3, Kimi K3 and DeepSeek V4.1 Flash returned nothing within 150–180 s, apparently queued on the free endpoints

Admitted to Dim's free worker pool after agent benchmarks: Qwen3.8-Flash via the Qoder CLI, and DeepSeek V4.1 Flash via the Letta Code CLI (about 3 parallel on the free plan)

Added to OpenCode's model catalogue (models.dev): Exo

Still answering (HTTP 200) after the Command Code channel ended

Calls return 404 "No endpoints found for stealth/space-bunny-alpha" (not retryable); re-checked 2/2 at 19:06. The model catalogue still lists it, so the catalogue is not a liveness signal.

"Space Bunny Alpha's stealth period has ended"; the model behind it to be revealed

Kilo's model name reads "Space Bunny Alpha (retires Oct 5)" (date we saw it, not when it was posted) → end date Oct 5

Announced as free on Freebuff on X: DeepSeek V4.1 Flash

Free limit raised to 300 requests/day

Shown as unavailable in the Zen catalogue

Announced as free on Freebuff on X: Solar Mini 4

"We just added 3 more 100% free models … Ling 3.1 Flash, Space Bunny Alpha, Laguna S 2.1"

Added to Kilo's model catalogue as "inclusionAI: Ling 3.1 Flash (new)" at $0 per token (Kilo's own record date)

Listed on OpenRouter as inclusionai/ling-3.1-flash at $0 (OpenRouter's catalogue record date)

Added to Cline's model list: Apodex 1.1 Mini

Added to the Zen model docs: Ling 3.1 Flash

Tencent notice of 2026-09-30 evening (carried by IT之家 2026-10-02): Hy3 free and Hy4 preview night free (23:00 to 08:00 next day) are both extended to 2026-10-31. Because participation was very high, the daily free quota is distributed per user and requests enter a queue when that day's capacity is busy (the page shows the reset time). Users who had never opened Hy4 preview had to start their first Hy4 preview chat before 2026-10-10 23:59 to get 14 days of daily free quota from that day → end date Oct 31

"Continues to be free … till Oct 5 on all plans" → end date Oct 5

Listed on OpenRouter as apodex/apodex-1.1-mini:free at $0 (OpenRouter's catalogue record date)

First seen in Kilo's model catalogue (models.dev sync): Apodex 1.1 Mini

Added to OpenCode's model catalogue (models.dev): Fledge Alpha

"The stealth period is now extended through October 5" → end date Oct 5

Made free in Command Code ("while it lasts")

Announced on the Vercel changelog: Ling 3.1 Flash

Announced as free on Freebuff on X: GPT-6.1 Sol

Ling 3.1 Flash launched on the official Ant Ling API, free for 2 weeks: Ling 3.1 Flash → end date Oct 14

Free on OpenCode "for two weeks"

Added to the Zen model docs: LongCat 2.5 Preview

Mentioned as free on NVIDIA build on X (earliest mention found): DeepSeek V4.1 Flash, GLM 5.3, Kimi K3

Announced as free on Command Code on X: Space Bunny (stealth)

Launched as a free stealth model (1M context; text, image and video input)

Free "for the next week" → end date Sep 30

First seen in Kilo's model catalogue (models.dev sync): Space Bunny (stealth)

Added to the Zen model docs: Space Bunny (stealth)

Mentioned as free on Freebuff on X (earliest mention found): MiMo V2.6 Flash, MiMo V2.6 Pro

Free "for the next week" → end date Sep 28

Added to the Zen model docs: JEV 1.13

Qoder international: Qwen3.8-Flash free for all personal users (billing 0.1× → 0.0×), first until Sep 30, then extended from Oct 1 with the end to be announced

Kimi K3 free in Cline for a limited time (Cline on X on Sep 20: "Kimi K3 is still free in the app for a limited time"); no end date

No longer listed in the OpenRouter catalogue capture of 2026-09-18T16:22Z; last priced $0 in the capture of 2026-09-17T00:59Z (approx end, an upper bound)

Mentioned as free on Freebuff on X (earliest mention found): GPT-6 Luna

Priced $0 in the OpenRouter catalogue capture of 2026-09-16T19:30Z (approx start; not listed in the capture of 2026-09-15T19:30Z)

Added to Kilo's model catalogue as "inclusionAI: Ling 3.0 Flash VL (free)" at $0 per token (Kilo's own record date)

Mentioned as free on Freebuff on X (earliest mention found): DeepSeek V4.1 Flash

Mentioned as free on SiliconFlow on X (earliest mention found): Qwen3.5 4B

Added to Kilo's model catalogue as "Nex AGI: Nex-N2.5-Pro (free)" at $0 per token (Kilo's own record date)

Added to Cline's model list: Ling 3.0 Flash Sante

Added to Kilo's model catalogue as "Dots Studio: Dots3-Note Preview (free)" at $0 per token (Kilo's own record date)

Added to Kilo's model catalogue as "MiniMax: MiniMax M2.7 (free)" at $0 per token (Kilo's own record date)

Announced as free on Freebuff on X: Solar Pro 4

Listed on OpenRouter as inclusionai/ling-3.0-flash-sante:free at $0 (OpenRouter's catalogue record date)

First seen in Kilo's model catalogue (models.dev sync): Ling 3.0 Flash Sante

Mentioned as free on Google AI Studio on X (earliest mention found): Gemini 3.8 Flash

Added to the Zen model docs: Muse Spark 1.3 (Contributor)

Added to Kilo's model catalogue as "MiniMax: MiniMax M3 (free)" at $0 per token (Kilo's own record date)

Added to the Zen model docs: Ling 3.0 Flash Fin

Announced as free on Freebuff on X: GLM 5.3 Flash

No longer listed in the OpenRouter catalogue capture of 2026-08-28T14:03Z; last priced $0 in the capture of 2026-08-25T12:44Z (approx end, an upper bound)

Mentioned as free on AIHubMix on X (earliest mention found): Coding Kimi K3

Added to Cline's model list: Inkling

First seen in Kilo's model catalogue (models.dev sync): Inkling Small

Priced $0 in the OpenRouter catalogue capture of 2026-08-21T15:02Z (approx start; not listed in the capture of 2026-08-14T06:05Z)

Added to the Zen model docs: Muse Spark 1.2 (Contributor)

Registered on ModelScope's free API-Inference gateway (model list timestamp): Qwen3.8 27B

Added to the Workers AI model catalogue: Qwen3.8 27B

Free tier back in the catalogue (HTTP 200)

Listed on OpenRouter as dots-studio/dots-3-note-preview:free at $0 (OpenRouter's catalogue record date)

Added to Cline's model list: Dots3-Note Preview

Mentioned as free on Google AI Studio on X (earliest mention found): Gemini 3.7 Flash

Guide published for using Qwen3.8 27B (free) through OpenRouter: Qwen3.8 27B

Letta: "DeepSeek V4 Flash are now available on Letta Cloud. Use your Letta API credits to run either model." Paid credits then, not free

Added to Cline's model list: LFM 2.5 2.6B

First seen in Kilo's model catalogue (models.dev sync): LFM 2.5 2.6B

Listed on OpenRouter as liquid/lfm-2.5-2.6b:free at $0 (OpenRouter's catalogue record date)

Listed on OpenRouter as nvidia/nemotron-3.5-lightning:free at $0 (OpenRouter's catalogue record date)

First seen in Kilo's model catalogue (models.dev sync): Nemotron 3.5 Lightning

Added to Cline's model list: Nemotron 3.5 Lightning

Added to the Zen model docs: Nemotron 3.5 Lightning

First seen in Kilo's model catalogue (models.dev sync): Laguna XS 2.1, North Mini Code, Step 3.7 Flash

DeepSeek V4 Flash updated in place to 0731 (re-post-trained on the same architecture); the API name deepseek-v4-flash is unchanged. Which snapshot each third-party channel serves is not recorded.

Listed on OpenRouter as thinkingmachines/inkling-small:free at $0 (OpenRouter's catalogue record date)

Added to Cline's model list: Laguna S 2.1

Added to Kilo's model catalogue as "Ling-3.0-flash (free)" at $0 per token (Kilo's own record date)

Added to Kilo's model catalogue as "Poolside: Laguna S 2.1 (free)" at $0 per token (Kilo's own record date)

Listed on OpenRouter as poolside/laguna-s-2.1:free at $0 (OpenRouter's catalogue record date)

Announced on the Vercel changelog: Laguna S 2.1

Listed on OpenRouter as thinkingmachines/inkling:free at $0 (OpenRouter's catalogue record date)

Added to Kilo's model catalogue as "Tencent: Hy3 (free)" at $0 per token (Kilo's own record date)

Added to Cline's model list: Laguna XS 2.1

Listed on OpenRouter as poolside/laguna-xs-2.1:free at $0 (OpenRouter's catalogue record date)

Inference experiment live; model table lists Qwen3.8-27B (docs Last change 2026-07-24)

Added to Cline's model list: North Mini Code

Listed on OpenRouter as cohere/north-mini-code:free at $0 (OpenRouter's catalogue record date)

Added to Kilo's model catalogue as "Nex AGI: Nex-N2-Pro (free)" at $0 per token (Kilo's own record date)

Added to Kilo's model catalogue as "NVIDIA: Nemotron 3 Ultra (free)" at $0 per token (Kilo's own record date)

Added to Cline's model list: Nemotron 3 Ultra (550B-A55B)

Added to the Zen model docs: Nemotron 3 Ultra (550B-A55B)

Listed as free on OpenRouter

Added to the Zen model docs: MiMo V2.5

Mentioned as free on NVIDIA build on X (earliest mention found): Nemotron 3 Super

Added to Kilo's model catalogue as "inclusionAI: Ring-2.6-1T (free)" at $0 per token (Kilo's own record date)

First seen in Kilo's model catalogue (models.dev sync): Nemotron 3 Nano Omni 30B A3B (reasoning)

Listed on OpenRouter as nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free at $0 (OpenRouter's catalogue record date)

Added to Cline's model list: Nemotron 3 Nano Omni 30B A3B (reasoning)

Added to Kilo's model catalogue as "Tencent: Hy3 preview (free)" at $0 per token (Kilo's own record date)

Added to Kilo's model catalogue as "inclusionAI: Ling-2.6-1T (free)" at $0 per token (Kilo's own record date)

Added to Kilo's model catalogue as "inclusionAI: Ling-2.6-flash (free)" at $0 per token (Kilo's own record date)

Added to Kilo's model catalogue as "Google: Gemma 4 26B A4B (free)" at $0 per token (Kilo's own record date)

Added to Cline's model list: Gemma 4 26B A4B, Gemma 4 31B

Listed on OpenRouter as google/gemma-4-26b-a4b-it:free at $0 (OpenRouter's catalogue record date)

Listed on OpenRouter as google/gemma-4-31b-it:free at $0 (OpenRouter's catalogue record date)

Added to Cline's model list: Nemotron 3 Super

No longer listed in the OpenRouter catalogue capture of 2026-03-19T16:24Z; last priced $0 in the capture of 2026-03-17T07:07Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2026-03-19T16:24Z; last priced $0 in the capture of 2026-03-17T07:07Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2026-03-17T07:07Z (approx start; not listed in the capture of 2026-03-06T15:49Z)

Priced $0 in the OpenRouter catalogue capture of 2026-03-17T07:07Z (approx start; not listed in the capture of 2026-03-06T15:49Z)

Added to Kilo's model catalogue as "NVIDIA: Nemotron 3 Super (free)" at $0 per token (Kilo's own record date)

Listed on OpenRouter as nvidia/nemotron-3-super-120b-a12b:free at $0 (OpenRouter's catalogue record date)

No longer listed in the OpenRouter catalogue capture of 2026-03-03T07:52Z; last priced $0 in the capture of 2026-02-19T04:56Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2026-02-15T04:12Z (approx start; not listed in the capture of 2026-02-01T23:22Z)

No longer listed in the OpenRouter catalogue capture of 2026-02-01T23:22Z; last priced $0 in the capture of 2026-01-28T22:57Z (approx end, an upper bound)

Launched on Z.ai: GLM-4.7-Flash

Launched free on BigModel: GLM-4.7-Flash

No longer listed in the OpenRouter catalogue capture of 2025-12-06T22:41Z; last priced $0 in the capture of 2025-11-30T07:53Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2025-11-30T07:53Z; last priced $0 in the capture of 2025-11-19T13:07Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2025-11-30T07:53Z; last priced $0 in the capture of 2025-11-19T13:07Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2025-11-30T07:53Z (approx start; not listed in the capture of 2025-11-19T13:07Z)

Priced $0 in the OpenRouter catalogue capture of 2025-11-19T13:07Z (approx start; not listed in the capture of 2025-11-14T18:31Z)

Priced $0 in the OpenRouter catalogue capture of 2025-11-19T13:07Z (approx start; not listed in the capture of 2025-11-14T18:31Z)

No longer listed in the OpenRouter catalogue capture of 2025-11-14T18:31Z; last priced $0 in the capture of 2025-11-08T12:17Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2025-11-07T01:40Z (approx start; not listed in the capture of 2025-11-02T18:55Z)

No longer listed in the OpenRouter catalogue capture of 2025-10-31T18:00Z; last priced $0 in the capture of 2025-10-22T18:39Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2025-10-22T18:39Z (approx start; not listed in the capture of 2025-10-20T00:18Z)

Added to OpenCode's model catalogue (models.dev): Big Pickle

No longer listed in the OpenRouter catalogue capture of 2025-09-22T19:04Z; last priced $0 in the capture of 2025-09-18T18:54Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2025-09-11T10:37Z (approx start; not listed in the capture of 2025-09-04T19:23Z)

Mentioned as free on SiliconFlow on X (earliest mention found): Qwen3 8B

Listed as free on Z.ai's pricing page (archived snapshot): GLM-4.5-Flash

Mentioned as free on SiliconFlow on X (earliest mention found): GLM-4 9B 0414

No longer listed in the OpenRouter catalogue capture of 2025-07-14T17:33Z; last priced $0 in the capture of 2025-07-14T08:26Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2025-07-04T15:40Z (approx start; not listed in the capture of 2025-07-01T11:11Z)

Google shut down gemini-2.5-pro-exp-03-25 (end of free access is at or before this date)

No longer listed in the OpenRouter catalogue capture of 2025-05-09T02:50Z; last priced $0 in the capture of 2025-05-05T17:01Z (approx end, an upper bound)

First public mention: GLM-4-Flash-250414

OpenRouter retired the free alpha endpoint (reveal post)

OpenRouter retired the free alpha endpoint (reveal post)

Priced $0 in the OpenRouter catalogue capture of 2025-04-10T18:22Z (approx start; not listed in the capture of 2025-04-03T04:31Z)

Priced $0 in the OpenRouter catalogue capture of 2025-04-10T18:22Z (approx start; not listed in the capture of 2025-04-03T04:31Z)

No longer listed in the OpenRouter catalogue capture of 2025-04-10T18:22Z; last priced $0 in the capture of 2025-04-03T04:31Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2025-04-10T18:22Z; last priced $0 in the capture of 2025-04-03T04:31Z (approx end, an upper bound)

Google changelog says gemini-2.5-pro-exp-03-25 stays usable on the free tier

OpenRouter announced free access to google/gemini-2.0-flash-exp:free (heavily rate limited)

No longer listed in the OpenRouter catalogue capture of 2024-11-22T22:53Z; last priced $0 in the capture of 2024-11-09T02:49Z (approx end, an upper bound)

Mentioned as free on SiliconFlow on X (earliest mention found): Qwen2.5 7B Instruct

No longer listed in the OpenRouter catalogue capture of 2024-09-05T18:09Z; last priced $0 in the capture of 2024-06-26T13:26Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2024-09-05T18:09Z; last priced $0 in the capture of 2024-06-26T13:26Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2024-09-05T18:09Z; last priced $0 in the capture of 2024-06-26T13:26Z (approx end, an upper bound)

No longer listed in the OpenRouter catalogue capture of 2024-06-19T08:19Z; last priced $0 in the capture of 2024-05-08T19:32Z (approx end, an upper bound)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2024-03-26T21:18Z (approx start; not listed in the capture of 2023-12-26T21:28Z)

Priced $0 in the OpenRouter catalogue capture of 2023-12-12T11:43Z (approx start; not listed in the capture of 2023-12-03T05:13Z)