Inkling (Thinking Machines Lab)
A Tier · 8.0/10
Mira Murati's $12B lab ships its first model (2026-07-15): a 975B/41B-active open-weights MoE that reasons natively over text, images, and audio with a 1M-token context -- positioned not as the strongest model, but as the best starting point for fine-tuning via Tinker
Score Breakdown
The Good and the Bad
What we like
- +First US frontier-adjacent open-weights release at this scale in 2026 -- 975B total / 41B active MoE (256 routed + 2 shared experts, 6 active per token, DeepSeek-V3-style routing), pretrained on 45T tokens
- +Natively multimodal INPUT across text, images, and audio (dMel spectrograms for audio, 40x40 pixel patches for vision) with a context window up to 1M tokens
- +Refreshingly honest positioning -- the launch post says outright it is 'not the strongest overall model available today, open or closed'; the pitch is calibration, an adjustable 'thinking effort' dial, and customizability
- +Designed for fine-tuning from day one: Tinker integration, cookbooks, and NVFP4 checkpoints lower the barrier to making it YOUR model rather than consuming it as-is
- +Credible team: Thinking Machines Lab is ex-OpenAI CTO Mira Murati's ~200-person lab, and this is its first public proof point after 18 months of infrastructure work
What could be better
- −Text-only outputs for now -- it reasons over images and audio but does not generate them
- −Not a benchmark leader by its own admission -- if you want maximum raw capability, frontier closed models (or Kimi K3 on the open side) outscore it
- −41B active parameters still puts practical self-hosting in multi-GPU territory -- this is not a laptop model (Inkling-Small's 12B active is the more realistic self-host target)
- −License terms live in the model card rather than the launch post -- verify commercial-use terms there before building on it
- −No first-party API; you depend on Tinker or third-party hosts (Together, Fireworks, Modal, Databricks, Baseten) for managed inference
Pricing
Self-hosted (Free)
- ✓Full weights on Hugging Face -- original checkpoint plus an NVFP4 checkpoint for efficient inference on NVIDIA Blackwell
- ✓Open weights; license terms per the model card (not spelled out in the launch post)
- ✓Inkling-Small (276B total / 12B active) available in preview
Tinker (fine-tuning platform)
- ✓Inkling available on Tinker day one with 64K and 256K context options
- ✓Fine-tuning with cookbook support, incl. three recipes for Inkling's audio capabilities
- ✓This is the business model: Inkling is the starting point, Tinker is how Thinking Machines makes money
Hosted APIs (partners)
- ✓Served by TogetherAI, Fireworks, Modal, Databricks (day-zero partner), and Baseten
- ✓No first-party inference API from Thinking Machines itself
System Requirements
Hardware needed to self-host. Min = smallest viable setup (usually heavy quantization). Max = full-precision / production-grade.
| Model variant | Min | Max |
|---|---|---|
| Inkling (975B total / 41B active MoE)Practically hosted-or-cluster for the full model; Inkling-Small (276B/12B active, preview) is the realistic self-host target | Multi-GPU server class -- 41B active parameters plus expert weights; NVFP4 checkpoint targets NVIDIA Blackwell systems | Production-grade multi-node inference or hosted via Together/Fireworks/Modal/Databricks/Baseten |
Known Issues
- LAUNCH (2026-07-15): Inkling is Thinking Machines Lab's first-ever model release -- open weights on Hugging Face (original + NVFP4 checkpoints), 975B total / 41B active MoE, native text+image+audio input, up to 1M context, pretrained on 45T tokens. **Inkling-Small** (276B total / 12B active) shipped alongside in preview, 'matching or exceeding its larger sibling on many benchmarks' per the vendor. Available on Tinker at launch (64K/256K context options, 50% launch discount) and via TogetherAI, Fireworks, Modal, Databricks, and Baseten. Strategy note: the company explicitly frames Inkling as a customization substrate monetized through Tinker fine-tuning, not a chatbot -- TechCrunch: revenue depends on the platform, not model distribution. Funding context: a reported ~$50B round was said to be coming together last November but had stalled by January (company declined comment)Source: Thinking Machines Lab (thinkingmachines.ai/news/introducing-inkling/), TechCrunch (2026-07-15), Reuters, Axios · 2026-07-15
- Vendor benchmark tables were not published as a simple comparative suite at launch, and the company itself disclaims SOTA -- treat any specific Inkling benchmark numbers circulating on aggregators as unverified until third-party evals (Artificial Analysis, LMArena) publishSource: thinkingmachines.ai launch post (no comparative benchmark table), TechCrunch · 2026-07-15
Best for
Teams that want to fine-tune a serious multimodal open-weights model into a domain specialist -- Tinker plus the cookbook ecosystem makes Inkling the most customization-oriented large release of 2026. Also researchers who want audio-native reasoning in an open model.
Not for
Anyone who just wants the smartest chat model out of the box (use Claude, GPT-5.6, or Kimi K3), or solo self-hosters without multi-GPU hardware -- watch Inkling-Small instead.
Our Verdict
Inkling is the most interesting kind of debut: a $12B lab's first model that deliberately refuses the benchmark arms race. The specs are real -- 975B/41B MoE, native text+image+audio input, 1M context, full weights on Hugging Face -- but the point is the Tinker fine-tuning flywheel: Inkling is sold as a starting ingredient, not a finished assistant. That makes it the leading US open-weights play (a lane Meta has drifted out of) and a genuinely different bet from Kimi K3's 'biggest model wins' approach launched the same week. If you consume models via API, this changes little today. If you build models from open weights, Inkling plus Tinker is now the most credible US-based path.
Sources
- Thinking Machines Lab: Introducing Inkling (2026-07-15) (accessed 2026-07-18)
- Thinking Machines Lab: Inkling product page (accessed 2026-07-18)
- TechCrunch: Thinking Machines amps up its bet against one-size-fits-all AI (2026-07-15) (accessed 2026-07-18)
Explore more Inkling (Thinking Machines Lab) rankings
Deeper leaderboards, benchmarks, task-specific tier lists, and status/pricing pages for Inkling (Thinking Machines Lab).
The Tier List Tuesday
Weekly newsletter: tier movers, new entrants, and the VS of the week. Built from our daily AI-tool sweeps. No spam, unsubscribe anytime.
Alternatives to Inkling (Thinking Machines Lab)
Llama 4 (Meta)
Meta's open-weights family -- Scout (10M context), Maverick (multimodal 400B MoE). NOTE: Meta's frontier work moved to the proprietary Muse Spark line in April 2026; Llama remains downloadable and supported but is effectively in maintenance mode
Mistral AI
**New 2026-10-15: Timestral, Mistral's first time-series forecasting family (5M, 28M, 123M and 209M parameters, weights on Hugging Face since 10/14 under a non-Apache licence), with the 209M model claimed first on GIFT-Eval, TIME and fev-bench and CPU inference in 10-140 ms.** **Mistral Large 4 ('le Chonk', ML4) entered public preview on 2026-10-06: a 1T-parameter natively multimodal MoE with roughly 50B active parameters, a 1M context window, callable today as mistral-large-4 at $1.36 / $0.14 cached / $4.18 per 1M tokens (batch at half price), open weights promised by the end of October, and vendor-claimed open-weight leads in cyber (Cybench 93%), coding (DeepSWE v1.1 61.7%) and agentic work (AutomationBench 59.9%).** **Mistral opened a Munich hub for Physics AI and Industrial AI on 2026-09-28**, confirming the May 2026 acquisition of Emmi AI (30+ physicists and engineers in computational fluid dynamics, structural mechanics and multi-physics simulation), naming BMW (crash simulation) and Siemens Energy as partners, and committing to **build one gigawatt of European compute capacity by 2030**. **Mistral now powers Firefox Smart Window (2026-09-16)** -- Mozilla's beta AI browsing assistant runs on Mistral models for France and North America, UK and Germany later this year, with zero data retention. **EUR 3B Series D at a EUR 21B+ post-money (9/08, Samsung-led)**; Cloudera partnership (9/10); Medium 3 retired 8/31 in favour of Medium 3.5.
DeepSeek
**DeepSeek-V4.1-Flash-1017 (2026-10-17) replaced V4.1-Flash on the web, app and API at unchanged prices** (the deepseek-flash model name now serves it; MIT weights on Hugging Face), with higher scores and shorter outputs, and DeepSeek opened its in-house **Search and Fetch APIs at $0.75 per 1,000 calls per endpoint** -- server-side Search/Fetch calls through DeepSeek Harness, Claude Code and Codex stay free until 2026-10-31 16:00 UTC. Earlier: **DeepSeek-V4.1-Flash shipped 2026-09-10**: a 552B MoE on a new Causal Encoder-Decoder architecture (8B active for input, 16B for output), natively multimodal, MIT-licensed on HF, and priced BELOW the V4-Flash it replaces -- $0.15/$0.60 off-peak, $0.30/$1.20 peak per 1M, with cache-hit input cut to $0.003/$0.006. V4-Flash and V4-Flash-Vision-Exp are retired (aliases route to V4.1-Flash). DeepSeek said V4-Pro would be phased out on 9/14, then **reversed that four days later** -- V4-Pro stays on the API at unchanged $0.66/$1.98 off-peak, $1.32/$3.96 peak until V4.1-Pro lands. V4.1-Flash's vendor benchmarks beat V4-Pro (DeepSWE 74.2, Terminal-Bench 2.1 90.6, GPQA 90.9)
Gemma 4 (Google)
Google DeepMind's open-weights model family -- multimodal, 256K context, runs on edge devices **EmbeddingGemma 2 (announced 2026-10-06; weights on Hugging Face since 9/14) adds a 740M-parameter Apache 2.0 embedding model built on Gemma 4 that maps text, code, images, video and audio into one 768-dimensional space, with an 8K context, Matryoshka truncation down to 128 dimensions and about 191 MB of RAM for text-only use on a Pixel 11 Pro.**
DiffusionGemma (Google)
Google DeepMind's experimental open-weights TEXT-DIFFUSION model (June 10, 2026) -- 26B MoE (3.8B active), Apache 2.0, generates 256-token blocks in parallel with bidirectional attention for up to 4x faster output (1,000+ tok/s on H100). Trades some quality vs Gemma 4 for raw speed
Qwen (Alibaba)
**Qwen-Image-2.1 went open-weight (weights 2026-09-14, blog 2026-09-20)** -- a unified text-to-image and editing model with a 7B visual-generation component, native transparent (RGBA) output, up to 10 reference images and mask-guided local edits -- but under a **research-only licence, not Apache**: commercial use needs a separate licence from Alibaba. Alibaba's open-weights + API family, and August 2026 was its biggest open-release month yet: **Qwen3.8-2.4T-A95B -- the Max-class flagship -- went open-weight on 2026-08-08** (custom Qwen3.8-Max licence), **Qwen3.8-27B** dense landed Apache 2.0 (8/05), and **Qwen3.8-Flash-Next** (8/24 weights, 8/26 blog) previews the Qwen4 architecture: 125B main + 51B n-gram embeddings with only 6B active, served as Qwen3.8-Flash at $0.15/$0.47 per 1M. Qwen 3.7 Max remains the GA API flagship ($2.50/$7.50)
GLM / Z.ai (Zhipu AI)
**GLM 5.2 has a named production deployment at Perplexity: on 2026-09-21 Perplexity disclosed that Perplexity Computer sessions are served by GLM 5.2, post-trained by Perplexity on real user sessions** (21.2% relative cut in live tool-call failures). **GLM-5.3 and GLM-5.3-Flash released 2026-08-25** (weights on Hugging Face): 5.3 is a post-training-only upgrade of the 5.2 base that Z.ai calls 'the most capable open-weights model for coding' (Terminal Bench 2.1 88.2, DeepSWE 66.9, CyberGym 84.5) but moves from MIT to a custom licence; 5.3-Flash (320B/18B active, multimodal) stays MIT at $0.15/$0.50 per 1M.
Kimi K3 (Moonshot)
Moonshot's 2.8T-parameter Kimi K3 (launched 2026-07-16/17) is the largest open-weight model ever released -- 1M context, multimodal, $3/$15 per 1M via API, ranked best-available on Arena.AI at launch. WEIGHTS SHIPPED ~2026-07-26/27 on Hugging Face (2.8T total / 104B activated, safetensors) under a custom Kimi K3 License, not the Modified MIT of the K2 line
Nemotron (Nvidia)
Nvidia's open-weights family -- hybrid Mamba-Transformer MoE architecture, optimized for efficient reasoning on Nvidia hardware. Nemotron 3 Ultra (550B total / 55B active) shipped 2026-06-04 as the family flagship, joining Super (120B/12B, March) and Nano
MiniMax M3
**MiniMax-H3's weights are public on Hugging Face (repo 2026-07-28, updated 2026-08-13) under a community licence whose open-weight grant is limited to the US, EU, UK and South Korea**, and **MiniMax Music 3 (2026-08-07)** ships open weights for five-minute full-song generation with a prominent-attribution commercial clause. MiniMax's coding/agent flagship -- M3 (June 1 2026): 1M-token context, MSA sparse attention (>15x decoding speedup at long context), SWE-Bench Pro 59.0%, Terminal-Bench 66.0%. OPEN WEIGHTS LIVE on HuggingFace since June 12 (~428B total / ~23B active, native multimodal, minimax-community license)
Falcon (TII)
UAE's Technology Innovation Institute open-weights family -- Falcon 3 optimized for efficient sub-10B deployment on consumer hardware
gpt-oss (OpenAI)
OpenAI's FIRST open-weight models -- gpt-oss-120b (single 80GB GPU, near parity with o4-mini on reasoning) and gpt-oss-20b (runs on 16GB edge devices). Apache 2.0. Launched 2025-08-05. gpt-oss-safeguard ships in 2026 as the safety-tuned variant
IBM Granite 4.0
IBM's enterprise-focused open-weight family -- Granite 4.0 hybrid Mamba-2 + transformer architecture (70-80% memory reduction vs pure transformer), 3B to 32B sizes, Apache 2.0. First open model family to secure ISO 42001 certification. Nano 350M runs on CPU with 8-16GB RAM. 3B Vision variant landed 2026-04-01
Arcee Trinity-Large-Thinking
Arcee AI's US-made open-weight frontier reasoning model -- launched 2026-04-01. 398B total params, ~13B active. Sparse MoE (256 experts, 4 active = 1.56% routing). Apache 2.0, trained from scratch. #2 on PinchBench trailing only Claude 3.5 Opus. ~96% cheaper than Opus-4.6 on agentic tasks
Olmo 3 (AI2)
Allen Institute for AI's fully-open frontier reasoning models -- Olmo 3 family (2025-11-20) includes 7B and 32B sizes, four variants (Base, Think, Instruct, RLZero). Apache 2.0 with fully open data + checkpoints + training logs. Olmo 3-Think 32B matches Qwen3-32B-Thinking at 6x fewer training tokens
AI21 Jamba2
AI21 Labs' hybrid SSM-Transformer (Mamba-style) open-weight family -- Jamba2 launched 2026-01-08. Two sizes: 3B dense (runs on phones / laptops) and Jamba2 Mini MoE (12B active / 52B total). Apache 2.0, 256K context, mid-trained on 500B tokens
StepFun Step 3.7 Flash
StepFun's (China) agent-focused open-weight family -- Step 3.7 Flash (May 28 2026): 198B sparse MoE vision-language model, ~11B active, 256K context, Apache 2.0, ~400 tok/s, SWE-Bench Pro 56.3. Supersedes Step 3.5 Flash (Feb 2026) as the flagship
Cohere Command A
Cohere's enterprise-multilingual flagship -- 111B params, 256K context, runs on 2x H100. 23 languages. CC-BY-NC 4.0 on weights (research / non-commercial), commercial requires Cohere enterprise contract. Follow-ups: Command A Reasoning + Command A Vision
LongCat-2.0 (Meituan)
Meituan's open-source 1.6T-parameter MoE (~48B active) with native 1M-token context, MIT license -- trained entirely on domestic Chinese AI ASICs and revealed as the stealth 'Owl Alpha' model that had been topping OpenRouter
Bonsai 27B (PrismML)
The first 27B-class model that runs on a phone (2026-07-14) -- ternary and 1-bit quantizations of Qwen3.6 27B squeeze a multimodal, tool-calling, 262K-context model into 3.9-5.9GB under Apache 2.0