# TensorFeed

> Follow the AI world: the latest AI news from ten publishers, whether OpenAI, Claude, Gemini and 20 other AI services are up right now, and the prices, benchmark scores and retirement dates of 80 AI models. Start a workflow for each new article or when a service goes down or recovers. No account or API key required.

App · 13 actions. Web page: https://docs.lodol.com/docs/integrations/tensorfeed

## Actions

### Get a model's price history

Get an AI model's price for each of the last one to seven days and whether it changed, to catch a price cut or rise. Prices are US dollars per million tokens.

**Inputs**

- Model (`model`, text, required): The model. Pick one, or give its id (claude-opus-5-5) or name (Claude Opus 5.5) from an earlier step.
- Days (`days`, number): How many days of prices to return, 1 to 7 (today included). 7 by default. From 1 to 7. Default: `7`.

**Outputs**

- `average_price_change_percent` (number): How much the average of the input and output price moved over the period, in percent.
- `change_count` (number): How many times the price moved.
- `days` (number): How many days are covered.
- `from_date` (text): The first day, YYYY-MM-DD.
- `has_history` (true/false): False when TensorFeed hasn't recorded prices for the model yet, such as one added today.
- `latest_input_price_per_million` (number): The latest input price.
- `latest_output_price_per_million` (number): The latest output price.
- `model_id` (text): The model's id.
- `model_name` (text): The model's name.
- `price_changed` (true/false): True when the input or output price moved in the period.
- `prices` (list): One entry a day, oldest first.
- `provider` (text): The company that makes it.
- `to_date` (text): The last day (today), YYYY-MM-DD.

**Example output**

```json
{
  "average_price_change_percent": 0,
  "change_count": 0,
  "days": 7,
  "from_date": "2026-10-13",
  "has_history": true,
  "latest_input_price_per_million": 4,
  "latest_output_price_per_million": 20,
  "model_id": "claude-opus-5-5",
  "model_name": "Claude Opus 5.5",
  "price_changed": false,
  "prices": [
    {
      "average_price_per_million": 12,
      "date": "2026-10-19",
      "input_price_per_million": 4,
      "output_price_per_million": 20
    }
  ],
  "provider": "Anthropic",
  "to_date": "2026-10-19"
}
```

### Get an AI model

Get one AI model's prices, context window, capabilities and release month, with its scores on the benchmarks TensorFeed tracks and who measured each one.

**Inputs**

- Model (`model`, text, required): The model. Pick one, or give its id (claude-opus-5-5) or name (Claude Opus 5.5) from an earlier step.

**Outputs**

- `benchmark_scores` (list): Its benchmark scores. Empty when TensorFeed has none for it.
- `currency` (text): Always USD.
- `model` (object): One AI model and its prices.
- `prices_updated_on` (text): When TensorFeed last updated its prices, YYYY-MM-DD.

**Example output**

```json
{
  "benchmark_scores": [
    {
      "benchmark": "Terminal-Bench 4.0",
      "benchmark_id": "terminal_bench_4",
      "independent": true,
      "max_score": 100,
      "note": "Vals, mini-swe-agent harness, avg@3, max effort.",
      "reported_by": "Vals AI",
      "score": 65.15,
      "source_url": "https://www.vals.ai/benchmarks/terminal-bench-4"
    }
  ],
  "currency": "USD",
  "model": {
    "capabilities": [
      "text",
      "vision",
      "tool-use",
      "code",
      "reasoning"
    ],
    "context_window": 1000000,
    "id": "claude-opus-5-5",
    "input_price_per_million": 4,
    "intelligence_benchmark_set": "2026",
    "intelligence_score": 97.1,
    "license": null,
    "name": "Claude Opus 5.5",
    "open_source": false,
    "output_price_per_million": 20,
    "provider": "Anthropic",
    "released": "2026-09",
    "tier": "flagship"
  },
  "prices_updated_on": "2026-10-19"
}
```

### Get an AI service's status

Check whether one AI service, such as the OpenAI API or Claude, is working right now, and which of its parts have a problem. Use it before a step that calls that service, to wait or switch providers when it is down.

**Inputs**

- AI service (`service`, text, required): The AI service, such as OpenAI API or Claude API.

**Outputs**

- `checked_at` (text): When TensorFeed last checked, ISO 8601 in UTC.
- `service` (object): One AI service's status.

**Example output**

```json
{
  "checked_at": "2026-10-19T22:36:02Z",
  "service": {
    "components": [
      {
        "name": "Chat Completions",
        "status": "operational",
        "status_label": "Operational"
      },
      {
        "name": "Responses",
        "status": "degraded",
        "status_label": "Degraded performance"
      }
    ],
    "is_operational": false,
    "last_checked_at": "2026-10-19T22:34:14Z",
    "name": "OpenAI API",
    "problem_components": [
      "Responses"
    ],
    "provider": "OpenAI",
    "status": "degraded",
    "status_label": "Degraded performance",
    "status_page_url": "https://status.openai.com"
  }
}
```

### List AI models

Compare AI models from Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek and more: price per million tokens, context window, what each can do and when it came out. Filter by maker, capability, tier or price, and sort by price or size to find the cheapest model that fits a job.

**Inputs**

- Made by (`provider`, text): Only models from this company. Leave empty for all of them.
- Can do (`capabilities`, list): Only models that can do everything ticked, such as reading images (vision) and calling tools. One or more of: `text`, `code`, `tool-use` (Tool use), `vision` (Vision (reads images)), `reasoning`, `audio`, `video`, `RAG` (Retrieval (RAG)).
- Tier (`tier`, text): Only models in this tier. Leave empty for every tier. One of: `flagship` (Flagship (most capable)), `mid` (Mid-range), `budget` (Budget (cheapest)).
- Only open-source models (`open_source_only`, true/false): Return only models whose weights are free to download. Off by default. Default: `false`.
- Highest input price (`max_input_price`, number): Only models that cost at most this many US dollars per million input tokens (a token is about three quarters of a word). At least 0.
- Order (`sort_by`, text): How to order the models. Newest first by default. One of: `newest` (Newest first), `input_price` (Lowest input price first), `output_price` (Lowest output price first), `context_window` (Largest context window first). Default: `newest`.
- Number of models (`limit`, number): The most models to return, 1 to 100. 25 by default. From 1 to 100. Default: `25`.

**Outputs**

- `count` (number): How many are in this result.
- `currency` (text): Always USD.
- `has_more` (true/false): True when more matched than the limit let through.
- `models` (list): The models, in the order asked for.
- `prices_updated_on` (text): When TensorFeed last updated its prices, YYYY-MM-DD.
- `sorted_by` (text): The order used: newest, input_price, output_price or context_window.
- `total_count` (number): How many matched in total, before the limit.

**Example output**

```json
{
  "count": 1,
  "currency": "USD",
  "has_more": true,
  "models": [
    {
      "capabilities": [
        "text",
        "vision",
        "tool-use",
        "code",
        "reasoning"
      ],
      "context_window": 1000000,
      "id": "claude-opus-5-5",
      "input_price_per_million": 4,
      "intelligence_benchmark_set": "2026",
      "intelligence_score": 97.1,
      "license": null,
      "name": "Claude Opus 5.5",
      "open_source": false,
      "output_price_per_million": 20,
      "provider": "Anthropic",
      "released": "2026-09",
      "tier": "flagship"
    }
  ],
  "prices_updated_on": "2026-10-19",
  "sorted_by": "newest",
  "total_count": 13
}
```

### List AI news

Get the latest AI news from TechCrunch, The Verge, WIRED, Hacker News, the NVIDIA, Google and Hugging Face blogs and more, newest first. Filter by category, publisher, a word, or articles published since a moment such as the last run. Covers the latest articles TensorFeed holds (about 100), updated every ten minutes.

**Inputs**

- Category (`category`, text): Only articles in this category, such as Research or Hardware/Chips. Leave empty for every category.
- Publisher (`source`, text): Only articles from this publisher, such as TechCrunch AI. Leave empty for every publisher.
- Mentioning (`contains`, text): Only articles whose headline or opening lines mention this word or phrase. Capitals are ignored.
- Published after (`published_after`, date and time): Only articles published after this moment, such as the time of the last run. Leave empty for all of them.
- Number of articles (`limit`, number): The most articles to return, newest first. TensorFeed holds about the latest 100. 20 by default. From 1 to 100. Default: `20`.

**Outputs**

- `articles` (list): The articles, newest first.
- `count` (number): How many are in this result.
- `feed_updated_at` (text): When TensorFeed last updated its news, ISO 8601 in UTC.
- `has_more` (true/false): True when more matched than the limit let through.
- `total_count` (number): How many matched in total, before the limit.

**Example output**

```json
{
  "articles": [
    {
      "categories": [
        "General AI"
      ],
      "id": "vddw2z",
      "published_at": "2026-10-19T22:04:28Z",
      "snippet": "Microsoft wants enterprises to closely watch the agents employees use with systems not controlled by AI labs like OpenAI and Anthropic.",
      "source": "TechCrunch AI",
      "source_domain": "techcrunch.com",
      "title": "Microsoft says AI agents are like insider threats as \u2018frontier models can\u2019t be trusted\u2019",
      "url": "https://techcrunch.com/2026/10/19/microsoft-says-ai-agents-are-like-insider-threats-as-frontier-models-cant-be-trusted/"
    }
  ],
  "count": 1,
  "feed_updated_at": "2026-10-19T22:36:00Z",
  "has_more": true,
  "total_count": 35
}
```

### List AI news from a past day

Get AI news from TensorFeed's saved copy of one day: for a daily digest of yesterday, or to look back at a launch. Returns up to 25 of the day's articles, which is all TensorFeed's free archive gives, and says how many it saved.

**Inputs**

- Date (`date`, date, required): The day to read, in UTC. TensorFeed's archive starts in May 2026.
- Number of articles (`limit`, number): The most articles to return, up to 25 (all TensorFeed's free archive gives for a day). 25 by default. From 1 to 25. Default: `25`.

**Outputs**

- `articles` (list): The day's articles.
- `count` (number): How many are in this result.
- `date` (text): The day read, YYYY-MM-DD.
- `has_more` (true/false): True when TensorFeed saved more that day than its free archive returns.
- `saved_at` (text): When TensorFeed saved the day's copy, ISO 8601 in UTC.
- `total_count` (number): How many articles TensorFeed saved that day.

**Example output**

```json
{
  "articles": [
    {
      "categories": [
        "General AI"
      ],
      "id": "vddw2z",
      "published_at": "2026-10-19T22:04:28Z",
      "snippet": "Microsoft wants enterprises to closely watch the agents employees use with systems not controlled by AI labs like OpenAI and Anthropic.",
      "source": "TechCrunch AI",
      "source_domain": "techcrunch.com",
      "title": "Microsoft says AI agents are like insider threats as \u2018frontier models can\u2019t be trusted\u2019",
      "url": "https://techcrunch.com/2026/10/19/microsoft-says-ai-agents-are-like-insider-threats-as-frontier-models-cant-be-trusted/"
    }
  ],
  "count": 1,
  "date": "2026-10-01",
  "has_more": true,
  "saved_at": "2026-10-01T23:50:15Z",
  "total_count": 125
}
```

### List AI podcast episodes

Get the latest episodes of AI podcasts such as Hard Fork, No Priors, Last Week in AI and The AI Daily Brief, newest first, with links to listen.

**Inputs**

- Podcast (`podcast`, text): Only episodes of this show, such as Hard Fork. Leave empty for every show.
- Mentioning (`contains`, text): Only episodes whose title or description mention this word or phrase. Capitals are ignored.
- Number of episodes (`limit`, number): The most episodes to return, newest first, 1 to 50. 10 by default. From 1 to 50. Default: `10`.

**Outputs**

- `count` (number): How many are in this result.
- `episodes` (list): The episodes, newest first.
- `has_more` (true/false): True when more matched than the limit let through.
- `total_count` (number): How many matched in total, before the limit.

**Example output**

```json
{
  "count": 1,
  "episodes": [
    {
      "audio_url": "https://anchor.fm/s/f7cac464/podcast/play/127738167/episode.mp3",
      "description": "Who gets to decide how AI wealth is shared?",
      "duration": "00:32:30",
      "duration_minutes": 32,
      "id": "48028h",
      "image_url": "https://d3t3ozftmdmh3i.cloudfront.net/podcast_uploaded_nologo/41472609.jpg",
      "podcast": "AI Daily Brief",
      "published_at": "2026-10-19T20:02:07Z",
      "title": "The AI Wealth Problem",
      "url": "https://podcasters.spotify.com/pod/show/nlw/episodes/The-AI-Wealth-Problem-e3qoorn"
    }
  ],
  "has_more": true,
  "total_count": 50
}
```

### List AI research papers

Get AI research papers from one of three daily lists: the newest on arXiv, Hugging Face's daily picks with their upvotes, or the most cited of all time. Each paper comes with its abstract, authors and link.

**Inputs**

- List (`paper_list`, text): Which list to read: the newest AI papers on arXiv, the papers Hugging Face's editors picked today, or the most cited AI papers of all time. Each is refreshed once a day. One of: `arxiv` (Newest on arXiv), `hugging_face` (Hugging Face daily picks), `most_cited` (Most cited (Semantic Scholar)). Default: `arxiv`.
- Mentioning (`contains`, text): Only papers whose title or abstract mention this word or phrase. Capitals are ignored.
- Number of papers (`limit`, number): The most papers to return, 1 to 50. 10 by default. From 1 to 50. Default: `10`.

**Outputs**

- `count` (number): How many are in this result.
- `has_more` (true/false): True when more matched than the limit let through.
- `list_label` (text): The list's name.
- `paper_list` (text): The list read: arxiv, hugging_face or most_cited.
- `papers` (list): The papers, in the list's order.
- `total_count` (number): How many matched in total, before the limit.
- `updated_at` (text): When TensorFeed last refreshed the list, ISO 8601 in UTC.

**Example output**

```json
{
  "count": 1,
  "has_more": true,
  "list_label": "Hugging Face daily picks",
  "paper_list": "hugging_face",
  "papers": [
    {
      "arxiv_id": "2610.20424",
      "authors": [
        "Enhan Zhao",
        "Yongqi Tong"
      ],
      "citation_count": null,
      "id": "2610.20424",
      "pdf_url": null,
      "published_at": "2026-10-14T00:00:00Z",
      "summary": "Existing long-horizon agent benchmarks often emphasize final task success.",
      "title": "BizBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business",
      "topic": null,
      "upvotes": 192,
      "url": "https://huggingface.co/papers/2610.20424",
      "venue": null,
      "year": null
    }
  ],
  "total_count": 30,
  "updated_at": "2026-10-19T14:15:15Z"
}
```

### List AI service incidents

Get the recent incidents on the AI services TensorFeed monitors (about the last day's), newest first: when each started and ended, how serious it was, a short summary and suggested next step. TensorFeed refreshes the list every two hours; for the live state use 'Get an AI service's status'.

**Inputs**

- AI service (`service`, text): Only this AI service. Leave empty for all of them.
- Only ongoing incidents (`ongoing_only`, true/false): Return only incidents that haven't been resolved. Off by default. Default: `false`.
- Number of incidents (`limit`, number): The most incidents to return, newest first, up to 25. From 1 to 25. Default: `25`.

**Outputs**

- `count` (number): How many are in this result.
- `has_more` (true/false): True when more matched than the limit let through.
- `incidents` (list): The incidents, newest first.
- `ongoing_count` (number): How many of the matching incidents are still open.
- `total_count` (number): How many matched in total, before the limit.
- `updated_at` (text): When TensorFeed last refreshed the list, ISO 8601 in UTC.

**Example output**

```json
{
  "count": 1,
  "has_more": false,
  "incidents": [
    {
      "affected_capabilities": [
        "inference"
      ],
      "duration_minutes": 68,
      "id": "Mistral-1792419858082",
      "impact": "informational",
      "ongoing": false,
      "provider": "Mistral AI",
      "recommended_action": "no_action",
      "recommended_action_label": "No action needed",
      "resolved_at": "2026-10-19T15:32:28Z",
      "service": "Mistral",
      "severity": "minor",
      "started_at": "2026-10-19T14:24:18Z",
      "summary": "Mistral experienced degraded performance for 68 minutes (14:24\u201315:32 UTC on 2026-10-19). Issue has been resolved.",
      "title": "Mistral degraded performance"
    }
  ],
  "ongoing_count": 0,
  "total_count": 1,
  "updated_at": "2026-10-19T22:15:14Z"
}
```

### List AI service statuses

Check whether the major AI services are working right now: OpenAI, Claude, Gemini, GitHub Copilot, Cursor, AWS Bedrock, Azure OpenAI, Mistral and more, 23 in all. TensorFeed reads each one's status page every two minutes. Turn on 'Only services with a problem' to get just the ones that aren't fully working.

**Inputs**

- Only services with a problem (`only_problems`, true/false): Return only the services that aren't fully operational right now. Off by default, which returns all of them. Default: `false`.

**Outputs**

- `all_operational` (true/false): True when every service is fully operational.
- `checked_at` (text): When TensorFeed last checked, ISO 8601 in UTC.
- `count` (number): How many are in this result.
- `operational_count` (number): How many are fully operational.
- `problem_count` (number): How many have a problem (degraded, an outage or maintenance).
- `services` (list): The services.
- `total_services` (number): How many services TensorFeed monitors.
- `unknown_count` (number): How many TensorFeed couldn't check just now.

**Example output**

```json
{
  "all_operational": false,
  "checked_at": "2026-10-19T22:36:02Z",
  "count": 2,
  "operational_count": 22,
  "problem_count": 1,
  "services": [
    {
      "components": [
        {
          "name": "Chat Completions",
          "status": "operational",
          "status_label": "Operational"
        },
        {
          "name": "Responses",
          "status": "degraded",
          "status_label": "Degraded performance"
        }
      ],
      "is_operational": false,
      "last_checked_at": "2026-10-19T22:34:14Z",
      "name": "OpenAI API",
      "problem_components": [
        "Responses"
      ],
      "provider": "OpenAI",
      "status": "degraded",
      "status_label": "Degraded performance",
      "status_page_url": "https://status.openai.com"
    },
    {
      "components": [
        {
          "name": "claude.ai",
          "status": "operational",
          "status_label": "Operational"
        },
        {
          "name": "Claude API (api.anthropic.com)",
          "status": "operational",
          "status_label": "Operational"
        }
      ],
      "is_operational": true,
      "last_checked_at": "2026-10-19T22:34:15Z",
      "name": "Claude API",
      "problem_components": [],
      "provider": "Anthropic",
      "status": "operational",
      "status_label": "Operational",
      "status_page_url": "https://status.anthropic.com"
    }
  ],
  "total_services": 23,
  "unknown_count": 0
}
```

### List model retirements

See which AI models OpenAI, Anthropic, Google, AWS Bedrock, Azure and others are retiring, when each stops working and what to switch to, soonest first. Pick dates to get, say, everything retiring in the next 90 days.

**Inputs**

- Provider (`provider`, text): Only retirements announced by this company or cloud, such as OpenAI or AWS Bedrock. Leave empty for all of them.
- Stage (`stage`, text): Only retirements at this stage. Leave empty for every stage. One of: `announced` (Retirement announced), `deprecated` (Deprecated (still works, being phased out)), `sunsetted` (Retired (no longer available)).
- Retiring on or after (`retiring_from`, date): Only models that stop working on or after this date. Pick today to see only what's still to come.
- Retiring on or before (`retiring_until`, date): Only models that stop working on or before this date.
- Model name contains (`model_contains`, text): Only models whose name or id contains this text, such as gpt-4.
- Number of retirements (`limit`, number): The most to return, soonest retirement first, 1 to 100. 25 by default. From 1 to 100. Default: `25`.

**Outputs**

- `count` (number): How many are in this result.
- `has_more` (true/false): True when more matched than the limit let through.
- `retirements` (list): The retirements, soonest first; those with no date yet come last.
- `total_count` (number): How many matched in total, before the limit.
- `updated_on` (text): When TensorFeed last updated the list, YYYY-MM-DD.

**Example output**

```json
{
  "count": 1,
  "has_more": true,
  "retirements": [
    {
      "announced_date": "2026-08-27",
      "days_until_retirement": 42,
      "deprecation_date": "2026-08-27",
      "id": "anthropic-claude-sonnet-4-5",
      "model": "claude-sonnet-4-5-20250929",
      "model_name": "Claude Sonnet 4.5",
      "notes": null,
      "provider": "Anthropic",
      "replacement": "claude-sonnet-5-5",
      "retirement_date": "2026-11-30",
      "source_url": "https://docs.anthropic.com/en/docs/about-claude/model-deprecations",
      "stage": "deprecated",
      "stage_label": "Deprecated (still works, being phased out)"
    }
  ],
  "total_count": 2,
  "updated_on": "2026-10-19"
}
```

### Rank AI services by uptime

Rank the AI services TensorFeed monitors by how reliably they worked over the last one to seven days, with the minutes each spent degraded or down. Pick a service to get just its figures.

**Inputs**

- Days (`days`, number): How many days to look back over, 1 to 7 (today included). 7 by default. From 1 to 7. Default: `7`.
- AI service (`service`, text): Only this AI service. Leave empty for all of them.

**Outputs**

- `check_interval_minutes` (number): How often TensorFeed checks each service, in minutes.
- `count` (number): How many are in this result.
- `days` (number): How many days are covered.
- `from_date` (text): The first day, YYYY-MM-DD.
- `services` (list): The services, most reliable first.
- `to_date` (text): The last day (today), YYYY-MM-DD.

**Example output**

```json
{
  "check_interval_minutes": 2,
  "count": 1,
  "days": 7,
  "from_date": "2026-10-13",
  "services": [
    {
      "checks": 5002,
      "degraded_checks": 374,
      "down_checks": 0,
      "downtime_minutes": 748,
      "hard_down_minutes": 0,
      "operational_checks": 4628,
      "rank": 21,
      "service": "OpenAI API",
      "unknown_checks": 0,
      "uptime_percent": 96.2615
    }
  ],
  "to_date": "2026-10-19"
}
```

### Rank models on a benchmark

Rank AI models by their score on one benchmark, such as SWE-bench for coding or GPQA Diamond for expert questions, best first. Each score says whether the model's maker or an independent tester reported it.

**Inputs**

- Benchmark (`benchmark`, text, required): The benchmark to rank by, such as SWE-bench for coding.
- Made by (`provider`, text): Only models from this company. Leave empty for all of them.
- Number of models (`limit`, number): The most models to return, best first, 1 to 100. 10 by default. From 1 to 100. Default: `10`.

**Outputs**

- `benchmark` (object): The benchmark.
- `count` (number): How many are in this result.
- `has_more` (true/false): True when more matched than the limit let through.
- `rankings` (list): The models, best first.
- `scores_updated_on` (text): When TensorFeed last updated its scores, YYYY-MM-DD.
- `total_count` (number): How many matched in total, before the limit.

**Example output**

```json
{
  "benchmark": {
    "description": "Real-world software engineering tasks from GitHub issues (SWE-bench Verified)",
    "id": "swe_bench",
    "max_score": 100,
    "name": "SWE-bench"
  },
  "count": 1,
  "has_more": true,
  "rankings": [
    {
      "independent": false,
      "model": "Claude Opus 5",
      "note": null,
      "provider": "Anthropic",
      "rank": 1,
      "released": "2026-07",
      "reported_by": "Anthropic",
      "score": 97,
      "source_url": "https://www.anthropic.com/claude-opus-5"
    }
  ],
  "scores_updated_on": "2026-10-19",
  "total_count": 47
}
```
