Mistral Medium 3 (72B Instruct) vs Together AI Inference-Time Compute API

“The primitive is clean: a permissively licensed, instruction-tuned 72B model you can run on two A100s and own outright. The DX bet is Apache 2.0 with no strings — no commercial restrictions, no model card carve-outs — which means you can actually build on this without a lawyer. The moment of truth is `huggingface-cli download mistralai/Mistral-Medium-3` and it works exactly as advertised. What earns the ship is the license decision, not the benchmark numbers — Mistral could have shipped this under a community-only license like Meta's earlier Llama terms and didn't, which is a genuine craft decision that respects the developer.”

“The primitive here is clean: wrap N parallel inference calls with a selection policy (majority vote or best-of-N scorer) and expose it as a single API parameter. That's the right abstraction — the complexity lives in the API layer, not in the caller's code. The DX bet is that developers shouldn't have to implement fan-out sampling logic themselves, and that bet is correct — running majority-vote naively means managing async calls, deduplication, and tie-breaking, which is annoying to get right. The specific technical decision that earns the ship: making N and the selection strategy first-class API parameters rather than a separate SDK or service layer means you can adopt this in one line of changed code, which is exactly where this kind of complexity should live.”

“Category is open-weight frontier models; direct competitors are Qwen2.5-72B-Instruct and Llama 3.3 70B — both strong, both Apache 2.0 or equivalent, both already deployed at scale. Mistral's coding and reasoning benchmark claims need scrutiny: they pick favorable evals and their leaderboard comparisons are author-curated, a pattern I flag every time. What actually earns a ship here is that Apache 2.0 at 72B is a real thing, self-hosting is straightforward, and the model is credibly competitive even if it isn't the undisputed winner the press release implies. What kills this in 12 months: Qwen3-72B or Llama 4's mid-tier already outperforms it and Mistral's API moat evaporates — the open weights survive but the commercial narrative doesn't.”

“Direct competitors are OpenAI's o-series with native best-of at the model level and self-hosted vLLM with sampling_n — both of which developers already use. What Together ships here is a managed version of a pattern that's well-understood, which is either obvious or genuinely useful depending on your infrastructure situation. Where this breaks: at high N values with long reasoning traces, costs multiply fast and latency becomes a product problem, not just an engineering one — and there's no mention of whether the scoring model for best-of-N is exposed or a black box. What kills this in 12 months: the major model providers ship native inference-time compute configuration that's tightly coupled to their own models, making provider-agnostic options less compelling. What earns the ship today: developers who want to apply this to open models without managing their own inference cluster have a real need that Together actually addresses.”

“The thesis: by 2027, most production LLM inference runs on self-hosted open-weight models, not API calls, because latency, cost, and data-residency requirements converge to make ownership mandatory for serious deployments. Mistral Medium 3 is a direct bet on that thesis — Apache 2.0 at a parameter count that fits on commodity enterprise GPU clusters (2x A100 80GB) puts self-hosting inside the reach of any mid-sized engineering team. The second-order effect that matters: Apache 2.0 at this capability tier accelerates the commoditization of the model layer, shifting power toward teams that own fine-tuning pipelines and proprietary data — the model becomes table stakes, the data flywheel becomes the moat. This tool is on-time to the open-weights consolidation trend, not early, but the Apache 2.0 decision is the specific variable that keeps it relevant.”

“The thesis here is falsifiable: scaling inference compute per query is a better return on investment than scaling training compute for reliability-sensitive tasks, and developers want that control surfaced at the API layer rather than baked into a specific model. The trend this rides is the inference-time scaling research that came out of 2024 — Together is early to productizing it as a generic API primitive rather than a model-specific feature, and that timing matters. The second-order effect that's underappreciated: once developers can dial accuracy vs. cost per request, they start building tiered products where cheap-and-fast handles 80% of queries and expensive-and-accurate handles the critical path — that's a new product architecture pattern, not just a performance knob. The future state where this is infrastructure: every serious LLM API offers inference-time compute budgeting as a standard parameter, and Together's head start on the API design shapes what that standard looks like.”

“The buyer for the weights is an engineer, not a budget holder — Apache 2.0 open weights don't generate revenue directly, and that's fine if the API business is the actual monetization story. The problem is the moat: Mistral's commercial API is competing against the same weights it just gave away, which means any customer doing sufficient volume will self-host and stop paying. The business survives only if Mistral's API offers something the raw weights don't — managed fine-tuning, guaranteed SLAs, enterprise contracts — and I don't see that story told clearly here. The specific thing that would flip this to a ship: a credible enterprise tier with switching costs baked into the workflow, not just the model.”

“The buyer is a developer or ML engineer at a company running accuracy-sensitive workloads — math tutoring, code generation, structured data extraction — and the budget comes from an AI infrastructure line. The pricing model is the problem: cost scales as N times the base token cost, which means the customers who get the most value are also the customers whose bills spike fastest, and there's no volume pricing or accuracy-based billing that aligns Together's revenue with customer success. The moat is thin — this is a sampling strategy layered on top of open models, and any inference provider can ship the same feature; Together's only defensible position is speed of iteration on open model support and pricing competitiveness. What would need to change for a ship: a pricing structure where Together captures a margin on the value of accuracy improvement rather than just multiplying the token cost, plus some proprietary scoring model for best-of-N that competitors can't trivially replicate.”