vendor evaluation

Snowflake Copilot is the wrong tool for executives querying in English. Cortex Analyst on Snowflake or Genie on Databricks wins — once you build the semantic layer; ThoughtSpot only earns its seat price across multiple warehouses.

OpenAI scored Astra for Law at 54% on Vals AI's private validation set, against its own GPT-6 Astra. On Vals' public leaderboard, three general frontier models already sit at 55.29% and Astra for Law is not listed.

A 40-call black-box audit from USTC flagged 7 of 15 OpenAI-compatible GPT services for behaviour consistent with output-token inflation. The flags are not proof, but padded answers pass quality evals and invoice checks alike, so measure output tokens per task against the first-party API.

A 176-setting controlled ablation shows the coding-agent harness moves success and cost more than the gap between the models being compared — and that one tool-surface flag reverses which model wins. A model swap is not a drop-in.

Only Lakera is a detector; NeMo Guardrails and Guardrails AI are frameworks that run detectors you still pick, host and tune. Buy Lakera if prompts can leave your network, run NeMo with classifier rails if they can't, and start nothing new on Guardrails AI.

A model router sells two things: mechanical failover, which works, and a learned classifier that picks your model, which the neutral benchmarks say does not. Buy the first, build the second, and plan on 35% savings rather than 60%.

The Era by Eon Benchmark ran nine models over 33 questions three times each and published the 95% intervals alongside the leaderboard. They are about 30 points wide, and 33 of the 36 pairwise comparisons could not be called. Your internal bake-off has the same problem.

ERPBench ran 100 identical ERP problems through six model families twice — solo against fixed opponents, then all six competing in one market. DeepSeek won the first, Gemini the second, and the two agreed on 21 of 100 tasks.

JPMorganChase compared 62 retrieval configurations on a production financial-news QA system for about $800, by judging the union of retrieved documents once and reusing 79.6% of the labels. Its top five embedding configurations spanned 0.007 MAP.

A study of 14,922 agent trajectories across five coding benchmarks finds Claude succeeds by matching the human patch's file scope while Qwen succeeds by exceeding it at every scale. Equal pass rates buy unequal diffs, and your ticket template is the control.

Four vendors each claim #1 on the same independent benchmark, on four different dates. Greptile holds the highest published precision at 76.2%, but research puts the gap between benchmark scores and real pull requests at 92%.

Two studies published the same day measured what grounds AI software recommendations. Perplexity's third-most-cited domain is a demo-software vendor's marketing blog, 59.8% of its citations sit below Tranco rank 100,000, and 34.7% of numeric citations point at a page that won't open or doesn't contain the number.

Microsoft now bundles Copilot Search free with a $30 Copilot seat and reaches over 100 connectors, so Glean has to beat it on the remainder rather than in the abstract. Dust has 12 connectors and replaces source-system ACLs with its own workspace spaces.

StartupBench ran nine frontier agents across 97 workflows taken from products AI startups already sell. A 73.67 rubric average produced an acceptable deliverable on 29.55% of runs — and agents satisfied auxiliary requirements more often than core ones.

Four agent orchestration products shut down, were superseded or repriced in twelve months. A six-criterion scorecard that weights exit cost at 30%, a six-week pilot with a portability test in week three, and the vendor landscape mapped to both.

Seven agent monitoring tools priced against the same workload: 50,000 runs a month at 12 steps each. Self-hosted Langfuse wins on cost and audit depth — but only one of these tools sits in the request path where it can actually stop a runaway agent, and that is the part you have to build yourself.

An arXiv benchmark of eleven production voice-agent stacks ran 3,300 simulated calls against a real database. The researchers found task completion collapses on complicated calls, per-call cost varies thirteenfold with no link to quality, and a verification leak hides inside correct refusals.