Benchmark
Zapier's AutomationBench, a benchmark that scores AI agents on multistep business workflows across apps and tools, used to compare frontier models.
The quarter-size global KV cache and eighth-size persistent storage are serving-cost claims rather than benchmark scores, and banking the second one requires a prefix-cache tier your stack has to already run.
Perspective Coverage
6 publishersReality
H Company's open-weight Holo4 27B handles GUIs, code, MCP servers and REST APIs in one model, scoring 61.7% on OSWorld 2.0 at an estimated $1.22 per task. The choice between specialists happens during training, so agent teams could drop the router from their stack if the scores hold on their own workloads.
Reality
OpenAI lists GPT-6.1 Sol at $2 per million input tokens and $10 per million output; the 10-cent figure in early coverage is its cached-input rate. Teams moving work off Astra should budget on the list rates and OpenAI's per-task costs.
Perspective Coverage
5 publishersReality
OpenAI's GPT-6.1 Sol halves cached-input pricing to $0.10 per million tokens and leaves standard rates at $2 and $10. Agents that resend long context collect the saving, while other buyers weigh gains shown mostly in OpenAI's own tests.
Reality
Meta hired CJ Desai, MongoDB's CEO of about 11 months, to run Meta Enterprise Platform, a new unit selling its Muse agent and AI stack to businesses. So far a buyer can weigh Desai's enterprise record and a product list, with no prices or contract terms.
Perspective Coverage
4 publishersReality
US companies are buying the cheapest AI model that can finish a job, the FT reports, with Ramp data pointing to a 41% fall in effective token prices. Investors now have to value AI vendors on cost per completed task, a figure that shifts with the workload.
Reality
OpenAI priced GPT-6 Sol at $2/$10 and Luna at $0.10/$0.50 per million tokens on September 22, half its GPT-5.6 promotional rates. Moving a job from Sol to Luna cuts its token rate by 95%, a bigger saving than the halving for any team whose work Luna can handle.
Reality
Anthropic took 20% off Opus 5.5 and OpenAI halved its two new GPT-6 tiers the same day. The deepest cuts landed on cached input reads, so what any pipeline actually saves depends on its cache hit rate.
Perspective Coverage
5 publishersReality
OpenAI says better caching and inference let it cut API prices for Sol and Luna by half, and the cost advantage it claims for the cheap tier over the old top tier comes in at one tenth on the benchmark it published and one hundredth in its summary.
Perspective Coverage
8 publishersReality
Anthropic and OpenAI shipped cheaper model tiers minutes apart on Tuesday. OpenAI halved Sol's posted token prices, and the only per-task cost comparison between the two labs so far comes from OpenAI itself.
Perspective Coverage
16 publishersReality
OpenAI's own benchmarks put GPT-6 Sol and Luna at half the list price and a fraction of a rival's cost per finished task. For defenders, the thing getting cheaper is autonomous tool calls into business systems.
Reality
Every token price on Claude Opus 5.5 fell 20% except cache reads, which dropped 60% to 20 cents a million. For the advertised 40% saving to come from price alone, half a buyer's Opus 5 bill has to be cache reads.
Reality
Sol now bills $2 and $10 per million tokens and Luna $0.10 and $0.50, while OpenAI quotes its own benchmark results per task, where the cheap model lands 2.2 points behind Sol on the software engineering test.
Reality
Two agentic post-training jobs, a 1.02-trillion-parameter model and a 309-billion cousin, ran with step time, reward and infrastructure error rates on the open web. Both stopped near step 30, and both models are still unreleased.
Reality
Sol lists at $2 and $10 per million tokens and Luna at $0.10 and $0.50. The per-task savings OpenAI published come mostly from the lower price, and the cheaper tier scores below its predecessor on computer use.
Reality
Z.ai's 320-billion-parameter model activates 18 billion per token and ships under MIT, so a buyer can download it and measure for themselves. Every capability figure published so far comes from Z.ai's own launch materials.
Reality
The rate undercuts the cached-input prices OpenAI and Anthropic publish by more than 130 times, and DeepSeek's own release says the sparse-attention design behind it has untested limits at cache boundaries.
Reality
Publishing weights and publishing something a team can actually deploy are different acts, and N2.5 lands on both sides of that line depending on which tier you pick and whether its weights exist yet.
Reality
The MIT-licensed 320B model card claims it beats GLM-5.2 at a tenth of the price and approaches Claude Opus 4.8 on coding, but it names no dollar rate, and the comparisons are largely the vendor's own.
Reality
Z.ai says the model runs at a tenth the cost of its last one. The comparison an operator needs is against the API invoice they already pay, and the release does not make it.
Reality
Invest · August 16, 2026 · Confirmed2 publishers