the AI bench
VERIFIED JULY 2026
All models

MODEL · INTERNSCIENCE (SHANGHAI AI LAB) · 35B TOTAL / 3B ACTIVE (MOE FLAGSHIP) · 4.5B DENSE (AGENTS-A1-4B)

Agents-A1 (35B-A3B + 4B)

An agent-specialist family post-trained on Qwen3.5 bases for long-horizon search, engineering, scientific research, instruction following, and tool calling. The pitch is agent-horizon scaling rather than parameter scaling: InternScience reports the 35B-A3B matching much larger models on agentic benchmarks, and the 4B — released July 14 — posting large gains over its own Qwen3.5-4B base (BrowseComp 66.8 vs 47.2, MatTools 49.3 vs 10.9). Unusually for a lab release, the evaluation framework is open-sourced so the numbers are reproducible.

License: Apache 2.0 · Context: 262K · Released: June 26, 2026 (35B-A3B); July 14, 2026 (4B)

The decision in five lines

The call
Consider — runnable locally, family reference
Best for
Local evaluation and family reference
Runs on
23 hardware picks fit (cheapest: Intel Arc B580 12 GB · $249)
Watch out
General chat or coding — this is an agent specialist, not a daily driver (the system prompt itself tells the model to answer directly without tools for ordinary questions).
Evidence
Estimated · last verified July 2026

35B total
PARAMETERS
AGENTIC SPECIALIST
TYPE
262K
CONTEXT
~20 GB (35B-A3B) / ~3 GB (4B) — the 4B runs on almost anything
VRAM AT Q4

Where we recommend this

This model isn’t currently in an active planner slot. See the runner notes below if you’re running it anyway.

The call

An agent-specialist family post-trained on Qwen3.5 bases for long-horizon search, engineering, scientific research, instruction following, and tool calling. The pitch is agent-horizon scaling rather than parameter scaling: InternScience reports the 35B-A3B matching much larger models on agentic benchmarks, and the 4B — released July 14 — posting large gains over its own Qwen3.5-4B base (BrowseComp 66.8 vs 47.2, MatTools 49.3 vs 10.9). Unusually for a lab release, the evaluation framework is open-sourced so the numbers are reproducible.

When not to use: General chat or coding — this is an agent specialist, not a daily driver (the system prompt itself tells the model to answer directly without tools for ordinary questions). All headline numbers are vendor-run, and the "trillion-parameter performance" framing is marketing: these are post-trains of Qwen3.5, not from-scratch models. Verify on your own agent loop before displacing a proven pick.

Runner notes

Official GGUFs (Q4_K_M / Q8_0 / F16) and FP8 from InternScience itself, plus mlx-community quants for Apple Silicon — so llama.cpp, Ollama, and LM Studio all work day one. vLLM/SGLang serve both sizes at 262K context on a single GPU; pass `--tool-call-parser qwen3_coder` for tool use, and `--language-model-only` on vLLM to skip the vision encoder and free KV-cache memory.

License
Apache 2.0
Released
June 26, 2026 (35B-A3B); July 14, 2026 (4B)
Maker
InternScience (Shanghai AI Lab)

Hardware that fits

Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.

Next step

Find-by-model — see what hardware runs this