Batch inference model catalog
Every model we target, including those we cannot serve yet. Nothing is listed as available until it is validated on its production hardware.
Target models available in the pilot: 0 / 6
| Model | Status | Execution | Pinned revision | What it still needs |
|---|---|---|---|---|
DeepSeek V4 Flashdeepseek-ai/DeepSeek-V4-Flashpublic checkpoint candidate; the provider recipe is not verified | Planned | Native only (optimization not applicable to this architecture) | 60d8d70770c6 | needs a multi-GPU host (none authorized); encoder parity test with the official encoding not run |
Llama 3.3 70B Instructmeta-llama/Llama-3.3-70B-Instructpublic base checkpoint; the "Fast" recipe is NOT established and is not offered | Validated on GPU, not yet offered | Native only (optimization not established) | 6f6073b42301 | no price set; no hosted capacity. Validated topology is a tight fit (8k context); the 4x80 GB profile with 32k context is not run |
Qwen3.6 27B (text-only)Qwen/Qwen3.6-27Bpublic checkpoint candidate | Validated on GPU, not yet offered | Native only (optimization not applicable to this architecture) | 6a9e13bd6fc8 | no price set; no hosted capacity. The 1x80 GB profile is not run (validated topology: 4x40 GB) |
Qwen3 4BQwen/Qwen3-4Bpublic base checkpoint; the "Fast" performance claim is not inherited | Validated on GPU, not yet offered | Native; an optimized profile runs on GPU, its quality scope is not yet published | 1cfa9a720891 | no price set; no hosted capacity (validated on 1x A100-SXM4-40GB; Metal profile not run) |
Qwen3 8BQwen/Qwen3-8Bbase-family candidate only; this is the ordinary public model, NOT a safety classifier | Validated on GPU, not yet offered | Native; an optimized profile runs on GPU, its quality scope is not yet published | b968826d9c46 | no price set; no hosted capacity |
Qwen3 8B with ReducerKV (50% prompt KV retention)Qwen/Qwen3-8B:reducerkv-r500 | Validated on GPU, not yet offered | Optimized (lossy KV-cache retention); runs on GPU, its quality scope is not yet published | b968826d9c46 | quality scope for customer acceptance not written: greedy outputs differed from native on 14 of 24 rows (quality not measured); no price set |
Qwen3 4B with ReducerKV (50% prompt KV retention)Qwen/Qwen3-4B:reducerkv-r500 | Validated on GPU, not yet offered | Optimized (lossy KV-cache retention); runs on GPU, its quality scope is not yet published | 1cfa9a720891 | quality scope for customer acceptance not written: greedy outputs differed from native on 20 of 24 rows (quality not measured); no price set |
Qwen3 32B with ReducerKV (50% prompt KV retention)Qwen/Qwen3-32B:reducerkv-r500 | Validated on GPU, not yet offered | Optimized (lossy KV-cache retention); runs on GPU, its quality scope is not yet published | 9216db5781bf | quality scope for customer acceptance not written: greedy outputs differed from native on 23 of 24 rows (quality not measured); no price set |
Qwen3 32BQwen/Qwen3-32Bbase-family candidate only; NOT the safety recipe and NOT the advertised AWQ variant (no quantized checkpoint id is invented) | Validated on GPU, not yet offered | Native; an optimized profile runs on GPU, its quality scope is not yet published | 9216db5781bf | no price set; no hosted capacity. The 2x80 GB TP=2 profile is not run (validated topologies: 4x40 GB TP=4; 1x80 GB at 8k context) |
Qwen3 0.6B (engineering reference, not on the target roster)Qwen/Qwen3-0.6B | Runs in local engineering tests only | Native only (optimization refused by the runtime) | c1899de289a0 | none for the local sample path; it is not a target-roster offering and Metal is not CUDA |
Qwen3.5 0.8B (remote serverless path, not on the target roster)Qwen/Qwen3.5-0.8B | Planned | Native only (optimization not applicable to this architecture) | 2fc06364715b | no price set; no hosted control plane (the accepted run used a loopback service on a Mac). The state stays ARTIFACT_VERIFIED: a remote acceptance is not GPU_VALIDATED, because the server configuration and the loaded weights are declared and readable in the provider worker log, not verified by the service. Whether to add a state for it is the owner's decision |
Video generation (reference catalog beta)deferred/video-generation-betarecorded only; different serving interface | Blocked | Native only (optimization not applicable to this architecture) | — | outside the text batch release (deferred, not silently forced through /v1/chat/completions) |
Robot action / policy (reference catalog beta)deferred/action-policy-betarecorded only; different serving interface | Blocked | Native only (optimization not applicable to this architecture) | — | outside the text batch release; robot-policy execution is out of scope |
Models are listed under their public identity. Labels such as Fast or Safety used by other providers describe their own recipes; we do not claim to reproduce them.
Video and robot-action models are outside this text batch service.
catalog 5d13bc85f47a · 2026-10-08