vs
SambaNova vs Inception
Both sell fast tokens. SambaNova gets speed from its own dataflow chip on large open models; Inception gets it from diffusion models on standard GPUs.
By The Subconscious Team · Updated
SambaNova vs Inception: key differences
SambaNova serves open models like MiniMax M2.7, DeepSeek and GPT-OSS 120B on its RDU chips, with GPT-OSS 120B at $0.22 in and $0.59 out and MiniMax M2.7 at about 800 tokens per second on its SN50 rack in SemiAnalysis tests, on hardware still ramping. Inception's Mercury 2.5 is a closed diffusion reasoning model with tool use and 260K context at $0.20 in and $0.75 out. Inception reports over 1,100 tokens per second; Artificial Analysis measures about 640.
SambaNova is the pick for fast decode on large open models, where quality runs higher than Mercury's cost-optimized tier. Inception is the pick for fast, cheap calls in autocomplete, voice and chat, and sells through AWS Bedrock and Azure Foundry. Both lean on vendor benchmarks for headline speed.
What SambaNova and Inception do
SambaNova
SambaNova designs its own inference chip, the Reconfigurable Dataflow Unit, and sells fast tokens on large open models through SambaCloud. The RDU maps the model graph onto the chip to cut trips to off-chip memory. A three-tier memory design of SRAM, HBM and bulk DRAM lets one system host very large models and hot swap between several of them in milliseconds. SambaCloud serves models like MiniMax M3 (in preview with a 1M context), MiniMax M2.7, DeepSeek, Gemma 4 31B and GPT-OSS 120B, with speeds reported by Artificial Analysis.
Example models: MiniMax M3, MiniMax M2.7, GPT-OSS 120B
Full SambaNova profile
Inception
Inception builds diffusion large language models (dLLMs). Every other LLM on this list generates text one token at a time; Inception's Mercury models start from a rough draft of a whole block and refine it over a few passes, producing many tokens in parallel. That ties speed less to output length and lets the models run fast on ordinary NVIDIA GPUs rather than custom chips. The company was founded by professors from Stanford, UCLA and Cornell behind core work on diffusion, Flash Attention and DPO, and is led by Stanford's Stefano Ermon. It raised a $50M seed in November 2025 led by Menlo Ventures, with Mayfield, M12, NVentures, Snowflake and Databricks, plus Andrew Ng and Andrej Karpathy as angels.
Example models: Mercury 2.5, Mercury 2
Full Inception profileShould you choose SambaNova or Inception?
SambaNova vs Inception at a glance
| Attribute | SambaNova |  Inception |
|---|---|---|
| Model access | Open weights | Closed weights (Mercury dLLMs) |
| Flagship models | MiniMax M3, MiniMax M2.7, GPT-OSS 120B | Mercury 2.5, Mercury 2 |
| Speed | ~800 tok/s on MiniMax M2.7 (SN50, SemiAnalysis) | 1,100+ tok/s (vendor); ~640 measured |
| Price | $0.22 in, $0.59 out (GPT-OSS 120B) | $0.20 in, $0.75 out (Mercury 2.5) |
| Customization | Unknown | Fine-tuning on request |
| Deployment | SambaCloud, racks for neoclouds | Serverless API, Bedrock, Azure, private |
| Long context | Up to 1M (MiniMax M3, preview) | 260K |
Frequently asked questions
What is the difference between SambaNova and Inception?
Both sell fast tokens. SambaNova gets speed from its own dataflow chip on large open models; Inception gets it from diffusion models on standard GPUs.
When should I choose SambaNova over Inception?
Fast decode on large open models; Agents that switch between models; Neoclouds adding a premium speed tier.
When should I choose Inception over SambaNova?
Very fast output on a small model; Voice agents and code autocomplete; Buying through Bedrock or Azure.
Is SambaNova or Inception cheaper?
SambaNova: $0.22 in, $0.59 out (GPT-OSS 120B). Inception: $0.20 in, $0.75 out (Mercury 2.5). The cheaper choice depends on the model and workload.
Which has more context, SambaNova or Inception?
SambaNova: Up to 1M (MiniMax M3, preview). Inception: 260K.
Related comparisons
Running long-horizon agents?
If your agents run past 200K tokens, compare both against Subconscious. Our inference stack treats a long-horizon trace as the primary workload, so speed, cost, and accuracy hold up deep into the trace.