# Inkling 2 Small (slim)

> Inkling 2 Small is Thinking Machines Lab's mixture-of-experts model, 594B total and 17B active parameters, taking text and images and returning text. It runs on Tinker for inference and fine-tuning, with open weights promised.

- Full: https://www.anchorterminal.com/tools/inkling-2-small.md (~7,900 tokens) · this version ~1,980 tokens · JSON https://www.anchorterminal.com/tools/inkling-2-small.json · canonical https://www.anchorterminal.com/tools/inkling-2-small
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-06 · pricing https://www.anchorterminal.com/tools/inkling-2-small#pricing · reviews https://www.anchorterminal.com/tools/inkling-2-small#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**D · 52.1/100 · rank #899 of 1354 · #28 in Model APIs & inference · not agent-ready · confidence medium** · same model listed separately: Tinker, SDK + MCP (https://www.anchorterminal.com/tools/tinker.min.md)


Assessment: A new multimodal mixture-of-experts model from Thinking Machines, callable today through Tinker's OpenAI-compatible endpoint at $0.30 and $1.00 per million tokens, with strong vendor-reported coding and agent scores. Tinker's serverless inference is beta, with no status page or SLA, and the weights aren't public yet.

## Facts

- Kind: Model API · vendor: Thinking Machines Lab · category: Model APIs & inference · legal entity: Thinking Machines Lab, Inc. (named as provider on the model card) · provenance 68/100
- Endpoint: `https://tinker.thinkingmachines.dev/services/tinker-prod/oai/api/v1` ()
- Auth: API key · pricing: Pay per use · x402: no · licence: Proprietary for now, used through Tinker under the Tinker Terms of Use and Thinking Machines' Model Acceptable Use Policy, per the model card. Thinking Machines says it will publish the weights under Apache 2.0 the week after launch, once outside safety testing is done
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The model as Tinker serves it, through serverless inference (beta) and the OpenAI-compatible API (beta), read from the launch post, the model card, Tinker's models and pricing page and its OpenAI-compatible API page on 6 November 2026. Platform-level readings (status, keys, audit log, terms, DPA, deprecations) are the tinker listing's of 1 October to 4 November. No account was made and nothing was called
- Model: A 48-layer decoder-only transformer with a sparse mixture of experts, 8 of 384 experts per token, hybrid local and global attention, 594B total and 17B active parameters. A 292M-parameter vision encoder based on Perception Encoder. MXFP8 and NVFP4 numerics. Text and images in, text out. No audio, which Inkling and Inkling-Small take
- Where it runs: Tinker's serverless inference (beta) at 256K context and its training and sampling at 64K or 256K, the Tinker Playground for chat, and the OpenAI-compatible endpoint, whose /chat/completions and /completions are both supported. Thinking Machines says other platforms will follow in the coming weeks
- Weights: Not public on 6 November 2026. The launch post says the full weights will be released under Apache 2.0 'next week after the completion of external safety testing and red-teaming', and that Tinker checkpoints become downloadable then. The model card's licence line is the Tinker Terms of Use with the Model Acceptable Use Policy
- Context: Up to 1M tokens per the model card, and evaluations used a 1M-token trajectory limit. Tinker serves 256K for serverless inference and 64K or 256K for training
- Pricing: Serverless inference $0.30 input, $0.03 cached input and $1.00 output per million tokens. Training side prefill $0.30, cached $0.03, sample $1.00, train $1.50 (64K) or $3.00 (256K). Inkling-Small's serverless prices are $0.30, $0.06 and $1.20, Inkling's $1.00, $0.17 and $4.05
- Thinking Machines' figures: Vendor-reported. AA Intelligence Index v5.0 30.6, Vals Index v2.2 56.1%, DeepSWE v1.1 72.1%, SWE-Bench Pro v2 Hard 86.3%, Vibe Code Bench v1.1 81.6%, FrontierCode v2.0 25.4%, Terminal Bench v4.0 13.1%, BrowseComp 90.2%, MMMU Pro 81.6%, HLE-Diamond 28.7%, Vals Finance Index 61.5%. Several rows are copied from Artificial Analysis, Vals, Scale AI and others by the publisher
- Safety: Pre-release testing for sycophancy, manipulation, psychological harm and over-refusal, with multi-turn outside red-teaming, CBRN and cyber testing including refusal-suppressed variants, and a loss-of-control assessment that puts it 'materially below frontier capabilities'. The launch post says it trails closed models such as GPT-6.1 Sol on alignment evaluations
- Data use: Tinker's Service Terms of Use (10 January 2026) commit not to use Customer Content for model training or fine-tuning, with a DPA and deletion on request, as read for the tinker listing

- Scores: Reliability 30, Performance pending, Schema & documentation 70, Agent ergonomics 62, Security & auth 57, Payments & pricing 20, Task success pending, Maintenance & community 75, Transparency & trust 64 · total over the 7 assessed categories
- Why: Reliability, Model reading, for the surface it's called on, Tinker's serverless inference and OpenAI-compatible API. · Schema & documentation, Model reading. · Agent ergonomics, Model reading (tool use, structured output, caching, context, batch, SDKs). · Security & auth, Platform readings are the tinker listing's. · Payments & pricing, Published rubric. · Maintenance & community, Model reading. · Transparency & trust, Editorial half.
- Sources: 8, open questions: 6, both in the full twin
- Capabilities: inference.llm
- JSON: https://www.anchorterminal.com/api/v1/tools/inkling-2-small.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/inkling-2-small.svg` or a link to https://www.anchorterminal.com/tools/inkling-2-small from a page on thinkingmachines.ai or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.
- Support: Discord https://discord.gg/KqqEZNX88c; tinker@thinkingmachines.ai; https://tinker-docs.thinkingmachines.ai/support/; https://github.com/thinking-machines-lab/tinker-feedback/issues; more in the full twin

## Before you call it

1. Call https://tinker.thinkingmachines.dev/services/tinker-prod/oai/api/v1 with an OpenAI client and TINKER_API_KEY, using /chat/completions
2. The serverless model ID is thinkingmachines/Inkling-2-Small:peft:262144:sampling-nvfp4, per Tinker's pricing page. The OpenAI page's examples use your own checkpoint paths, so confirm it before relying on it
3. Pass reasoning_effort as none to xhigh or 0.0 to 0.99 (default 0.9); reasoning arrives in reasoning_content unless separate_reasoning is false
4. Keep images between 40 and 4096 px a side, per the model card
5. Validate JSON yourself, since structured output and the tool-call format aren't documented
6. Plan for a move to open weights. When they publish, the open-weights listing will carry sizes and runtimes

## Connect

```bash
curl https://tinker.thinkingmachines.dev/services/tinker-prod/oai/api/v1/chat/completions \
  -H "Authorization: Bearer $TINKER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "thinkingmachines/Inkling-2-Small:peft:262144:sampling-nvfp4",
    "messages": [{"role": "user", "content": "Hello!"}],
    "reasoning_effort": "low"
  }'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| OpenAI API | A | 83.9 | inference.llm | https://www.anchorterminal.com/tools/openai-api.min.md |
| BlockRun.AI | A | 78.9 | inference.llm | https://www.anchorterminal.com/tools/blockrun-ai.min.md |
| Claude API | BB | 77.3 | inference.llm | https://www.anchorterminal.com/tools/anthropic-api.min.md |
| GroqCloud | BB | 74.5 | inference.llm | https://www.anchorterminal.com/tools/groq.min.md |
| Fireworks AI Inference | B | 69.3 | inference.llm | https://www.anchorterminal.com/tools/fireworks-inference.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)