Ling 3.0 Flash
inclusionAIReleased Jul 23, 2026Open weights · MITling-3.0-flash
- Reasoning
- yes
- Context
- 262K tokens
- Max output
- 32k tokens
- Input
- text
- API features
- tools, JSON mode, seed, logprobs
Read Nov 4, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from inclusionAI
The cheapest host, Novita AI, does not state its weights precision. Weights and context per host
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| Novita AIprecision not stated · 256K ctx · 32K out | $0.06 | $0.18 | $0.012 | — | — | — | Nov 4 |
| DeepInfrabf16 · 128K ctx (256K) · 32K out | $0.06 | $0.18 | $0.012 | 95 | 0.43 s | 96.5% | Nov 4 |
2 of 2 providers · prices per million tokens, read Nov 4, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Nov 4, 06:00 UTC · uptime: OpenRouter, last 24 h · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| Novita AI | not stated by the host | 256K | 32K | yes | no | yes |
| DeepInfra | bf16 | 128K | 32K | yes | no | yes |
Weights, context, max output: from the host where its terms allow, else as declared to OpenRouter (hover for source and date) · features as declared to OpenRouter, not verified by us
Ling 3.0 Flash spec sheet, field by field, with sources
Ling 3.0 Flash spec sheetmaker-declared
What inclusionAI declares, from its own weights and docs. What each host serves is in the offers: both can be true.
- Context window
- 262,144 tokens
- Parameters
- 127,486,405,600≈ 127.5B
- Active experts
- 8 of 512 active per token
- Weight precision
- bfloat16
- License
- MIT — open weights, permissiveScope: the license of the published weights only — not the maker's acceptable-use policy, nor the terms of the host you call.
- Knowledge cutoff
- not published by the maker
- Modalities
- text
- Max output (official)
- not published by the maker
- Release date
- not published by the maker
- Maker lifecycle status
- not published by the maker
Also declared in the weights
- Architecture class
- BailingMoeV3ForCausalLM
- Layers
- 42
- Vocabulary
- 157,184 tokens
- Weights first published
- August 2, 2026
- Weights last updated
- October 12, 2026
inclusionAI/Ling-3.0-flash · config.json @ fcc60a5 · read 2026-10-13 · 6 of 10 fields published by the maker
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| Novita AI | $0.15 | $0.57 | $0.42 |
| DeepInfra | $0.15 | $0.57 | $0.42 |
What each host bills for 1,000 requests of a representative job
Your own workload
- Parameters
- 124B
- 5.1B active per token (MoE)
- Memory needed
- 74 GB
- in INT4 · 149 GB in INT8
- Smallest cloud setup
- A100
- 80 GB · $1.19/h, about $869 a month
- Worth it above
- 4B tokens
- a month, against the cheapest API at $0.24 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 74 GB | A100 · $1.19/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 149 GB | 2× H100 · $5.28/h | near-lossless |
| FP16 (full) | 298 GB | 4× H100 · $10.56/h | reference quality |
At home: in INT4 it fits the Ryzen AI Max+ 395 (128 GB unified, ~96 usable) — about $3,487 once, or $0.1327/h over 3 years, electricity aside, at 256 GB/s of memory bandwidth — that, not FLOPS, sets tokens per second; it pays for itself after about 15B tokens against the cheapest API.
Memory ≈ parameters × bytes × 1.2 · cloud: on-demand, 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
Other inclusionAI models · all its models, hosts and history ›
- Ling 3.0 Flash VL (inclusionAI)2 providers · from $0.24
- Ling 3.1 Flash2 providers · from $0.30
- Ling 3.0 Flash Fin (inclusionAI)1 provider · from $0.29
- Ling 3.0 Flash Sante (inclusionAI)1 provider · from $0.29