Inkling
Inkling is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 41B active parameters out of 975B total. It is designed for general-purpose reasoning, coding, agentic and tool-use systems, retrieval-augmented generation, instruction following, and multilingual conversational applications. Its native image and audio understanding supports multimodal analysis alongside text.
Model card on Hugging Face- Input / output price
- $1.00 / $4.05 per 1M
- Context
- 1M tokens
- Released
- Jul 15, 2026
Paste into Claude Code, Cursor, or Codex
Help me set up the Respan gateway: fetch and follow https:/
Providers
3 providers, cheapest first, retired then OpenRouter last, priced per 1M tokens after gateway discounts.
| Provider | Input | Output |
|---|---|---|
Concentrate concentrate/inkling | $1.20/M | $5.00/M |
FireworksRetired fireworks/inkling | $1.00/M | $4.05/M |
OpenRouter openrouter/inkling | $1.00/M | $4.05/M |
Is a provider down? Fireworks AI status, OpenRouter status
Pricing
Averaged across 3 providers, in USD per 1M tokens after gateway discounts.
- Average input
- $1.07 per 1M
- Average output
- $4.37 per 1M
- Average cache read
- $0.20 per 1M
Cost by volume
Total cost from 0 to 1T tokens on the cheapest and most expensive provider.
Throughput and latency
30-day averages from requests through the Respan gateway. Under each figure, every model on the gateway by its average, this one in blue.
Output speed
Higher is better32tokens/sacross providers
Time to first token
Lower is better0.5sacross providers
Response latency
Lower is better19sacross providers
Benchmarks
The Respan Index plus 39 published scores from 11 sources, each linked to where it was published.
31.9/ 100
Respan Index#85 of 176
- 85Inkling31.9
- 84GLM-4.732.2
- 86Claude Sonnet 4.531.3
Published scores
| Benchmark | Score | Source | Published |
|---|---|---|---|
| SWE-bench Verifiedthinkingmachines.ai | 77.6% | thinkingmachines.ai | Jul 15, 2026 |
| LiveBench CodingLiveBench | 71% | LiveBench | Oct 5, 2026 |
| Terminal-Bench 2.1 (best harness)thinkingmachines.ai | 63.8% | thinkingmachines.ai | Jul 15, 2026 |
| SWE-bench Pro (public)thinkingmachines.ai | 54.3% | thinkingmachines.ai | Jul 15, 2026 |
| WeirdMLHåvard Tveit Ihle | 32.3% | Håvard Tveit Ihle | Oct 5, 2026 |
| FrontierSWE V2Proximal Labs | 4.1% | Proximal Labs | Oct 5, 2026 |
| Terminal-Bench 4.0Vals AI | 0.5% | Vals AI | Oct 5, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| BrowseComp (w/ ctx management)thinkingmachines.ai | 77.1% | thinkingmachines.ai | Jul 15, 2026 |
| MCP Atlasthinkingmachines.ai | 76% | thinkingmachines.ai | Jul 15, 2026 |
| LiveBench Agentic CodingLiveBench | 49.4% | LiveBench | Oct 5, 2026 |
| Toolathlon Verifiedthinkingmachines.ai | 45.5% | thinkingmachines.ai | Jul 15, 2026 |
| Toolathlon-VerifiedHKUST | 45.5% | HKUST | Jul 15, 2026 |
| APEX-AgentsMercor | 33.8% | Mercor | Oct 5, 2026 |
| tau2-bench Banking KnowledgeSierra | 25% | Sierra | Aug 4, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| GPQA DiamondEpoch AI | 88.3% | Epoch AI | Oct 5, 2026 |
| LiveBench ReasoningLiveBench | 78.3% | LiveBench | Oct 5, 2026 |
| LiveBench Data AnalysisLiveBench | 72.8% | LiveBench | Oct 5, 2026 |
| SimpleBenchSimpleBench | 50% | SimpleBench | Jul 22, 2026 |
| Chess PuzzlesEpoch AI | 21% | Epoch AI | Oct 5, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| AIME 2026thinkingmachines.ai | 97.1% | thinkingmachines.ai | Jul 15, 2026 |
| OTIS Mock AIME 2024-2025Epoch AI | 88.9% | Epoch AI | Oct 5, 2026 |
| LiveBench MathematicsLiveBench | 88.4% | LiveBench | Oct 5, 2026 |
| FrontierMath Tiers 1-3 v2Epoch AI | 33.3% | Epoch AI | Oct 5, 2026 |
| FrontierMath Tier 4 v2Epoch AI | 4.9% | Epoch AI | Oct 5, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| Global-MMLU-Litethinkingmachines.ai | 88.7% | thinkingmachines.ai | Jul 15, 2026 |
| IFBenchthinkingmachines.ai | 79.8% | thinkingmachines.ai | Jul 15, 2026 |
| LiveBench LanguageLiveBench | 73.5% | LiveBench | Oct 5, 2026 |
| LiveBench Instruction FollowingLiveBench | 70.1% | LiveBench | Oct 5, 2026 |
| SimpleQA Verifiedthinkingmachines.ai | 43.9% | thinkingmachines.ai | Jul 15, 2026 |
| SimpleQA VerifiedEpoch AI | 40.3% | Epoch AI | Oct 5, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| CharXiv RQ (with python)thinkingmachines.ai | 82% | thinkingmachines.ai | Jul 15, 2026 |
| CharXiv RQthinkingmachines.ai | 78.1% | thinkingmachines.ai | Jul 15, 2026 |
| MMAUthinkingmachines.ai | 77.2% | thinkingmachines.ai | Jul 15, 2026 |
| Audio MCthinkingmachines.ai | 56.6% | thinkingmachines.ai | Jul 15, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| LMArena EloLMArena | 1443 Elo · #75 of 191 | LMArena | Nov 19, 2026 |
| LMArena WebDevLMArena | 1414 Elo · #85 of 124 | LMArena | Nov 19, 2026 |
| LMArena AgentLMArena | -0.1845 · #66 of 68 | LMArena | Nov 18, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| VoiceBenchthinkingmachines.ai | 91.4% | thinkingmachines.ai | Jul 15, 2026 |
| SAGEVals AI | 36.6% | Vals AI | Oct 5, 2026 |
Compare Inkling: Inkling vs Inkling 2 Small, Inkling vs Inkling-Small
Apps
Use Inkling in the tools you already run. Each guide points the app at the Respan gateway.
Setup guides for more apps are in the integration docs.
Inkling FAQ
Short answers, built from the data on this page.
Inkling is a model from Thinking Machines, released on Jul 15, 2026. Inkling is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 41B active parameters out of 975B total. It is designed for general-purpose reasoning, coding, agentic and tool-use systems, retrieval-augmented generation, instruction following, and multilingual conversational applications. Its native image and audio understanding supports multimodal analysis alongside text.
On the Respan gateway, Inkling costs from $1.00 per 1M input tokens and $4.05 per 1M output tokens on Fireworks and OpenRouter, after gateway discounts. Averaged across its 3 providers, it costs $1.07 per 1M input tokens and $4.37 per 1M output tokens.
Inkling is available through 3 providers: Concentrate, Fireworks, and OpenRouter.
Inkling accepts up to 1M tokens of context on its largest provider route.
Point any OpenAI-compatible SDK at https://gateway.respan.ai/v1 with your Respan API key and set the model to "inkling". To set it up with a coding agent, copy the agent setup prompt at the top of this page.
Inkling scores 31.9 out of 100 on the Respan Index v1.1, ranking #85 of 176 models. By category: Coding 6.1, Agents 43.1, Reasoning 31.8, Knowledge & language 39.6, and Human preference 51.4. The Respan Index combines independent and lab-reported benchmark results into one score, using benchmark data collected by Respan.
Inkling has 39 published benchmark scores, including SWE-bench Verified 77.6%, BrowseComp (w/ ctx management) 77.1%, and GPQA Diamond 88.3%. Each score links to where it was published; the data is collected by Respan.
Certified where it counts.
Built to meet the security and privacy standards that enterprise and healthcare teams require.
ISO 27001
The internationally recognized standard for information security management.
SOC 2
Secure, compliant management of your data across all of our systems.
GDPR
Operated under GDPR, the world's strictest standard for data privacy.
HIPAA
HIPAA compliant, with a BAA available for healthcare teams.