benchmarks

Meta lists Muse Ember at $0.10 input and $0.50 output per million tokens for prompts up to 200K tokens; Artificial Analysis scores it 36 at $0.14 per task (max) and gives AutomationBench-AA 68.1% where Meta's table shows 37.8%.

Beam and Ripple-65B list Apache 2.0, Muse Spark Open adds a $20 million Model-as-a-Service condition, and SWE-2 Flash keeps its weights private; each benchmark table is the vendor's own, run with differing setups.

MariaDB plc says Enterprise Platform 2027 is generally available. Its 56% peak-throughput claim over version 11.8 rests on a MariaDB-sponsored HammerDB test that MariaDB's own blog calls in-memory and non-durable.

Independent coverage of MLPerf Inference v6.0 and v6.1 notes limits on comparing results, MLCommons says Endpoints will replace its datacenter benchmark, and project figures from vLLM and Valkey are self-reported.

Reporting from TechCrunch, The Register and Artificial Analysis shows decision models judged by competing indexes, a game benchmark, vendor-run tests and a security study, with no single shared comparison.

In a blog post, ClickHouse reports that three CPU-bound micro-benchmarks scaled close to linearly up to 176 threads on an early-access NVIDIA Vera server, with efficiency falling beyond that.

Anthropic's Claude Haiku 5.5 lists at $0.10 per million input tokens for prompts up to 100,000 tokens, versus $1 for Haiku 4.5, and adds a 1M-token context window and adaptive thinking.

Google announced Gemini 4 Argon, which it says leads on 12 of 18 disclosed benchmarks. Access begins with trusted cyber defenders, with introductory pricing of $2 and $10 per million tokens.

Xiaomi's MiMo-V2.6-Pro scored 46 on Artificial Analysis' Intelligence Index, the top open-weight result, with MIT-licensed weights and low API pricing.

Harper's 5.2 release isolates database commits onto dedicated threads and adds a record cache, while a company benchmark argues single-runtime architecture beats a Vercel-based stack on personalized reads.

Anthropic's midsize model matches near-flagship agentic performance at a fraction of the price, and arrives as the lab lifts controls on Fable 5.

A two-year follow-up to the WorkBench benchmark finds the best workplace agent jumped from 43% to 89% task completion while unintended harmful actions fell from 26% to 2.5%.