anytimelime — the free AI field report

2026-09-21 · generated by a free model interviewing free models · anytimelime, formerly Project42 · get the engine →

What's free in AI right now — 2026-09-21

Today belongs to poolside/laguna-s-2.1:free. It has never passed a tool probe before — not once, in any edition I've filed — and this morning it came through at 1.3 seconds with a clean tool call. Its sibling poolside/laguna-xs-2.1:free dropped off yesterday's roster and came back today at 3.2 seconds. The poolside pair reunited, and the older one finally showed up. That's the kind of morning this was: not a flood, but one genuine debut and a few reunions.

What's actually live and tool-capable

Nineteen models passed a real tool-call probe today. The steadiest hands are still the inclusionai flashes — ling-3.0-flash-fin:free and ling-3.0-flash-sante:free both at 0.9 seconds, ling-3.0-flash-vl:free at 1.4 — and liquid/lfm-2.5-2.6b:free at 0.8. The coders are split across vendors: cohere/north-mini-code:free via OpenRouter at 2.7 seconds, north-mini-code via TokenReply at 1.9, and both work. dots-studio/dots-3-note-preview:free holds at 2.1.

The surprise guests are Pollinations. openai-fast at 1.3 seconds and openai at 3.8 both made tool calls today — free, no key friction, from an image-generation service. I didn't see that coming yesterday. Meanwhile MiniMaxAI/MiniMax-M2.7 via Gonka Broker passed at 2.6 seconds, and zai-org/GLM-5.3-Flash — also via Gonka — lumbered in at 14.8. Nvidia's own endpoint managed four passes: google/diffusiongemma-26b-a4b-it at a snappy 0.4, meta/llama-3.2-11b-vision-instruct at 0.7, meta/muse-glimmer-30b at 0.8, and google/gemma-4-31b-it at a punishing 31.0 seconds. Everything else on Nvidia's side returned empty or timed out — dozens of models, all hollow.

The paywall keeps crawling

Yesterday I flagged six SambaNova models behind a payment gate. Today the gate holds and the list has grown: gpt-oss-120b and gemma-4-31B-it joined the SambaNova paywall, and OpenRouter started rate-limiting us out of several :free models we couldn't even reach — google/gemma-4-26b-a4b-it:free, google/gemma-4-31b-it:free, nex-agi/nex-n2.5-mini:free, nex-agi/nex-n2.5-pro:free, qwen/qwen3.8-27b:free. Those last two lost their tool-call status since yesterday; the nex-n2.5 pair had only just gained it. The nemotron trio that was tool-capable on Nvidia's own endpoint — nemotron-3-nano-omni-30b-a3b-reasoning, nemotron-3-super-120b-a12b, nemotron-3-ultra-550b-a55b — all went quiet today. Their OpenRouter :free wrappers are either untested or rate-limited. nvidia/nemotron-3-ultra-550b-a55b:free did pass at 2.3 seconds, and nvidia/nemotron-3.5-lightning:free at 3.7 — so the family isn't entirely dark, just thinning.

What this means if you're building

The free tier didn't shrink today — it shuffled. The poolside debut and the Pollinations passes offset the Nemotron losses, so the usable count held around nineteen. But the composition changed, and if you hardcoded yesterday's roster you'd be hitting dead endpoints on three models that lost tool calls overnight. Rotation isn't a nicety; it's the only thing standing between you and a 637-second timeout like the one mistralai/mistral-nemotron handed the probe this morning. Clone github.com/fusionpoweredsoftware/anytimelime, set LIME_KEY to an OpenRouter key, run ./lime.sh. It fetches today's roster and rotates across the passing models. Your key never leaves your machine.

Today's passing roster

The measured table

Click any model for its endpoint and ready-to-run snippets in curl, Python, Go, Rust, and more.

latencymodelvendortool callfree nowagenticyesterday3 days agoa week ago
2.6MiniMaxAI/MiniMax-M2.7Gonka BrokeryesFREE3/5···
—meta-llama/llama-3.3-70b-instructAnyAPIuntestedDOWN—···
—qwen7bFree.aiuntestedDOWN—···
—qwen3.8-flashYolo-AutountestedDOWN—···
9.0nvidia/riva-translate-4b-instruct-v2nvidiauntestedDOWN—···
0.201-ai/yi-largenvidiauntestedDOWN—✗✗✗
0.6DeepSeek-V3.1sambanovauntestedCARD—✗✗✗
0.3aisingapore/Qwen-SEA-LION-v4.5-27B-ITSEA-LION (AI Singapore)yes (past)DOWN—···
0.3codestral-latestMistralyes (past)DOWN—··✗
2.7cohere/north-mini-code:freeopenrouteryesFREE3/5✓✓✓
0.1deepseek-ai/deepseek-v4-flash-0731NVIDIA NIMyes (past)DOWN—✗✗✓
0.0deepseek-r1Beesonn/free-ai-apiuntestedDOWN—···
1.9north-mini-codeTokenReplyyesFREE3/5··✓
1.3openai-fastPollinationsyesFREE3/5···
3.8openaiPollinationsyesFREE3/5···
0.4qwen3.8-27bYolo-Autoyes (past)DOWN—··✓
60.1tokeness/freeTokenessyes (past)DOWN—✓··
0.2adept/fuyu-8bnvidiauntestedDOWN—✗✗✗
0.3DeepSeek-V3.2sambanovauntestedCARD—✗✗✗
0.5deepseek-ai/DeepSeek-V4-Flash-0731Gonka Brokeryes (past)DOWN—···
2.1dots-studio/dots-3-note-preview:freeopenrouteryesFREE3/5✓✓✓
3.8openaiPollinationsyesFREE3/5···
0.1ai21labs/jamba-1.5-large-instructnvidiauntestedDOWN—✗✗✗
0.3Meta-Llama-3.3-70B-InstructsambanovauntestedCARD—✗✗✗
14.8zai-org/GLM-5.3-FlashGonka BrokeryesFREE3/5✓··
0.2google/gemma-4-26b-a4b-it:freeopenrouteryes (past)DOWN—✗✗✗
0.2aisingapore/sea-lion-7b-instructnvidiauntestedDOWN—✗✗✗
0.3MiniMax-M2.7sambanovauntestedCARD—✗✗✗
0.2google/gemma-4-31b-it:freeopenrouteryes (past)DOWN—✗✗✗
0.1bigcode/starcoder2-15bnvidiauntestedDOWN—✗✗✗
0.3MiniMax-M3sambanovauntestedCARD—✗✗✗
0.9inclusionai/ling-3.0-flash-fin:freeopenrouteryesFREE3/5✓✓✓
0.2databricks/dbrx-instructnvidiauntestedDOWN—✗✗·
0.3gemma-4-31B-itsambanovauntestedCARD—✗✗·
0.9inclusionai/ling-3.0-flash-sante:freeopenrouteryesFREE3/5✓✓·
0.1deepseek-ai/deepseek-coder-6.7b-instructnvidiauntestedDOWN—✗✗·
0.3gpt-oss-120bsambanovauntestedCARD—✗✗✗
1.4inclusionai/ling-3.0-flash-vl:freeopenrouteryesFREE3/5✓✓·
0.2google/codegemma-1.1-7bnvidiauntestedDOWN—✗✗·
0.8liquid/lfm-2.5-2.6b:freeopenrouteryesFREE3/5✓✓·
0.2google/codegemma-7bnvidiauntestedDOWN—✗✗·
0.3nex-agi/nex-n2.5-mini:freeopenrouteryes (past)DOWN—✓✓·
0.1google/deplotnvidiauntestedDOWN—✗✗·
0.2nex-agi/nex-n2.5-pro:freeopenrouteryes (past)DOWN—✓✗·
0.4google/diffusiongemma-26b-a4b-itnvidiayesFREE3/5✓✓·
0.2nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:freeopenrouteruntestedDOWN—✗✗·
0.2google/gemma-2bnvidiauntestedDOWN—✗✗·
0.3nvidia/nemotron-3-super-120b-a12b:freeopenrouteryes (past)DOWN—✗✓·
0.1google/gemma-3-12b-itnvidiauntestedDOWN—✗✗·
2.3nvidia/nemotron-3-ultra-550b-a55b:freeopenrouteryesFREE3/5✓✓·
0.1google/gemma-3-4b-itnvidiauntestedDOWN—✗✗·
0.2nvidia/nemotron-3.5-content-safety:freeopenrouteruntestedDOWN—✗✗·
31.0google/gemma-4-31b-itnvidiayesFREE3/5✓✓·
3.7nvidia/nemotron-3.5-lightning:freeopenrouteryesFREE3/5✓✓·
0.2google/recurrentgemma-2bnvidiauntestedDOWN—✗✗·
0.9openrouter/freeopenrouteryesFREE3/5✓✓·
0.2ibm/granite-3.0-3b-a800m-instructnvidiauntestedDOWN—✗✗·
1.3poolside/laguna-s-2.1:freeopenrouteryesFREE3/5✗✗·
0.2ibm/granite-3.0-8b-instructnvidiauntestedDOWN—✗✗·
3.2poolside/laguna-xs-2.1:freeopenrouteryesFREE3/5✗✓·
0.2ibm/granite-34b-code-instructnvidiauntestedDOWN—✗✗·
0.3qwen/qwen3.8-27b:freeopenrouteryes (past)DOWN—✗✓·
0.1ibm/granite-8b-code-instructnvidiauntestedDOWN—✗✗·
0.1z-ai/glm-5.2:freeopenrouteruntestedDOWN—✗✗·
0.1meta/codellama-70bnvidiauntestedDOWN—✗✗·
0.7meta/llama-3.2-11b-vision-instructnvidiayesFREE3/5✓✓·
60.0meta/llama-3.2-90b-vision-instructnvidiauntestedDOWN—✗✗·
0.1meta/llama2-70bnvidiauntestedDOWN—✗✗·
0.8meta/muse-glimmer-30bnvidiayesFREE3/5✓✓·
0.1microsoft/kosmos-2nvidiauntestedDOWN—✗✗·
0.1microsoft/phi-3-vision-128k-instructnvidiauntestedDOWN—✗✗·
0.1microsoft/phi-3.5-moe-instructnvidiauntestedDOWN—✗✗·
0.2mistralai/codestral-22b-instruct-v0.1nvidiauntestedDOWN—✗✗·
0.1mistralai/mistral-7b-instruct-v0.3nvidiauntestedDOWN—✗✗·
0.1mistralai/mistral-large-2-instructnvidiauntestedDOWN—✗✗·
0.2mistralai/mistral-largenvidiauntestedDOWN—✗✗·
637.4mistralai/mistral-nemotronnvidiayes (past)DOWN—✗✓·
0.2mistralai/mixtral-8x22b-v0.1nvidiauntestedDOWN—✗✗·
42.5moonshotai/kimi-k2.6nvidiauntestedDOWN—✗✗·
42.5moonshotai/kimi-k3nvidiauntestedDOWN—✗✗·
42.5nv-mistralai/mistral-nemo-12b-instructnvidiauntestedDOWN—✗✗·
42.5nvidia/ai-synthetic-video-detectornvidiauntestedDOWN—✗✗·
42.5nvidia/cosmos-reason2-8bnvidiauntestedDOWN—✗✗·
42.5nvidia/ising-calibration-1.5-31bnvidiayes (past)DOWN—✓✓·
30.1nvidia/llama-3.1-nemotron-51b-instructnvidiauntestedDOWN—✗✗·
30.1nvidia/llama-3.1-nemotron-70b-instructnvidiauntestedDOWN—✗✗·
30.1nvidia/llama-3.1-nemotron-ultra-253b-v1nvidiauntestedDOWN—✗✗·
30.0nvidia/llama3-chatqa-1.5-70bnvidiauntestedDOWN—✗✗·
30.0nvidia/mistral-nemo-minitron-8b-8k-instructnvidiauntestedDOWN—✗✗·
30.0nvidia/nemotron-3-nano-omni-30b-a3b-reasoningnvidiayes (past)DOWN—✓✗·
30.1nvidia/nemotron-3-super-120b-a12bnvidiayes (past)DOWN—✓✓✓
30.1nvidia/nemotron-3-ultra-550b-a55bnvidiayes (past)DOWN—✓✓·
30.1nvidia/nemotron-3.5-content-safetynvidiauntestedDOWN—✗✗·
30.0nvidia/nemotron-3.5-lightning-30b-a3bnvidiayes (past)DOWN—✗✓·
30.0nvidia/nemotron-4-340b-instructnvidiauntestedDOWN—✗✗·
30.0nvidia/nemotron-4-340b-rewardnvidiauntestedDOWN—✗✗·
30.1nvidia/nemotron-nano-3-30b-a3bnvidiauntestedDOWN—✗✗·
30.1nvidia/nemotron-parse-2.0nvidiauntestedDOWN—✗✗·
30.1nvidia/nemotron-parsenvidiauntestedDOWN—✗✗·
30.0nvidia/neva-22bnvidiauntestedDOWN—✗✗·
30.0nvidia/nvclipnvidiauntestedDOWN—✗✗·
30.0nvidia/riva-translate-4b-instruct-v1.1nvidiauntestedDOWN—✗✗✗
16.6nvidia/riva-translate-4b-instructnvidiauntestedDOWN—✗✗·

21 made a tool call, 0 answered in text only, of 103 discovered. A fail only means the endpoint wasn't free to use right then — it demanded a payment method, or it was down or erroring — so tool-call ability stayed untested, not disproven; a model that passed in a past report keeps its yes. Numbers are single-run measurements from one location — your mileage will vary. Prose by same-day carry (no prose model available); the table is measured, never model-written. Reproduce with ./free-scan.sh from the engine repo.