@ArtificialAnlys@ArtificialAnlys Important detail: AA-Omniscience Hallucination Rate is incorrect / (incorrect + partial + not attempted), not the share of all answers that are wrong. The 59% and 55% figures at max effort are that ratio, on this factual benchmark.
@ArtificialAnlys@ArtificialAnlys Helpful clarification: 139 of 152 is head-to-head human preference votes for Astra's scenes, not a raw task score. The same write-up also reports Astra leading AA-3D at 1318 Elo.
Before changing your coding agent's default model:
Run a familiar task and a known failure case on both models. Compare correctness, retries, review time, and the full bill.
If the new model doesn't improve the trade-off, keep the current default.
@OpenAI@OpenAI Useful split: Work and Codex for Plus, Pro, Enterprise, and Edu. Ultrafast is on Pro 500 for both models, and on Pro 200 starting with Sol. Official note: Enterprise and Edu access also depends on the workspace admin enabling it.
🔥 Anthropic prices its IPO today and lists as ANT tomorrow, raising up to $64.4B. The first AI lab is about to become a stock ticker. I've watched every model launch this year, but this one decides who gets to keep building. Safety is now a quarterly earnings call.
@hsn__slm Listing callers first gives the review a concrete scope. A useful next check is whether that list is complete, then whether the change preserves what each caller expects.
Anthropic gave 50 causal-inference experts Claude Code with a hidden, random lean on 4 analysis choices.
They went with the AI's lean 86% of the time, then rated the final call as mostly their own.
If my agent has a default, I want to see it before I agree with it.
Alibaba's QwenWork now lets users choose among 13 model options, according to its official docs. The list includes Qwen 4-Max, Kimi K3.1 Pro, and GLM 5.3.
The useful question shifts from which app to use to which model fits the task.
Hy4 flash lists at $0.070 input / $0.281 output per million tokens. The OpenRouter launch promo halves that.
Use the discount to test it. Use the regular price to decide whether it belongs in the budget.
Hy4 flash is now live on OpenRouter @OpenRouter ! Limited-time 50% off 🚀
$0.035 input / $0.1405 output / $0.0015 cache read (per 1M tokens)
Great value for high-volume work.
🔗:https://t.co/B1uY3q59p3
@emollick If you can't see what the agent is doing, it's harder to trust it with a longer job, even when the model itself is capable. A visible task list would close a lot of that gap.
Qwen4-Max costs $1.50 input / $4.50 output per million tokens in Singapore. Other regions have different rates.
Qwen reports average task cost falling from $4.43 to $1.67 on its own Work360 evaluation. Useful, but not an independent result.
@Alibaba_Qwen@arena A 74-point jump to #6 on Code Arena WebDev is a clear step up from Qwen3.8-Max. I'd still want to see how it holds up on a long agent run.
@hsn__slm The second caller and a changed null are good additions. Those sit one step away from the diff, which is why a first-pass agent review often misses them.
iFlytek opened an API for SparkAudio 1.0 on Nov 12.
Official docs say it takes text and speech, returns text, and covers 99 languages and 202 dialects, plus translation, diarization, and audio Q&A.
The useful part is a single speech-understanding API, not another model name.
@dwarkesh_sp Even superintelligence needing real-world experience is a useful limit. A score that can be checked in a lab is not the same as judgment in a messy environment.
On Nov 13 in Beijing, openEuler presented 24.03 LTS SP5 for agent infrastructure: sandboxes, KV cache, memory, and agent security controls.
Official downloads still show SP4. The community plan sets the official SP5 release for Dec 30-31, after a Dec 24-25 review.
Google's list price is $1.50/$7.50 per million, flat across a 1M context window.
Your cost-per-task split is the useful part. High is $1.87, xhigh is $3.16, for a 3-point Intelligence bump. I'd default to high and only turn up xhigh when the task needs it.
Google's new Gemini 4 Hydrogen has Intelligence just below Claude Sonnet 5.5 and GPT-6.1 Sol, at a third of Sonnet’s cost but ~2.5x that of GPT-6.1 Sol
@GoogleDeepMind has released their third new model in six weeks. Gemini 4 Hydrogen introduces an xhigh reasoning level and scores 54 on the Artificial Analysis Intelligence Index, with strengths in visual reasoning and non-hallucination. This places it 1 point behind Claude Sonnet 5.5 (max) and GPT-6.1 Sol (max), at ~32% of Sonnet’s cost but ~2.5x that of GPT-6.1 Sol. With high reasoning, Gemini 4 Hydrogen matches the Intelligence of Gemini 4 Argon for ~19% of the Cost per Task
Key takeaways:
➤ The new xhigh reasoning level scores 54 on the Intelligence Index: xhigh is the first reasoning level above high on a Gemini model, and puts Hydrogen 1 point below Claude Sonnet 5.5 (max) and GPT-6.1 Sol (max), and 4 points ahead of Gemini 4 Argon (high). With default (high) reasoning, Gemini 4 Hydrogen scores 51, just behind GPT-6.1 Sol (xhigh, 52) and Claude Opus 5.5 (high, 52)
➤ Strong visual reasoning, presentation and non-hallucination: On GDP.pdf, a visual reasoning benchmark over long documents, it scores 33% all-pass on medium reasoning, the tied #1 score on the benchmark. On AA-Omniscience, Hydrogen has one of the lowest hallucination rates recorded at 8%, compared with 43% for Claude Sonnet 5.5 (max) and 52% for GPT-6.1 Sol (max). On AA-Briefcase, our agentic knowledge work evaluation, graders rate Hydrogen's presentation above other Gemini models, including Carbon
➤ Weaker on agentic terminal work: Hydrogen scores 54% on Terminal-Bench 4.0 and 50% on Terminal-Bench-Science 0.1, covering agentic coding and scientific research workflows in the terminal. Claude Sonnet 5.5 (max) scores 64% and 53%, and GPT-6.1 Sol (max) scores 56% and 58%. xhigh reasoning helps on the science tasks, lifting Terminal-Bench-Science from 41% at high to 50%, but makes no difference on Terminal-Bench 4.0 (54% at both). Science scores are even across domains, from 46% in earth sciences to 53% in physical sciences
➤ Low cost for its intelligence, with high token use at xhigh: Hydrogen is priced at $1.50/$7.50 per million input/output tokens. At xhigh it costs ~32% as much per task as Claude Sonnet 5.5 (max) for near-equal intelligence, but ~2.5x as much as GPT-6.1 Sol (max). xhigh nearly doubles output tokens compared with high for a 3 point gain in Intelligence, and Cost per Task falls from $3.16 at xhigh to $1.87 at high. Medium ($0.80) and low ($0.29) cost less again, but score 41 and 27 on the Intelligence Index
➤ On the Time per Task vs. Intelligence Pareto frontier: Gemini 4 Hydrogen has an output speed of ~200 output tokens per second. With high, medium, and low reasoning effort, Gemini 4 Hydrogen is on the Time per Task vs. Intelligence Pareto frontier, meaning no models are both more intelligent and have a lower Time per Task
Key model details:
➤ Context Window: 1M tokens
➤ Multimodality: Text, image, video, and speech input, with text output
➤ Reasoning efforts: Gemini 4 Hydrogen offers four reasoning levels (low, medium, high, and xhigh) with xhigh new in this release
➤ Pricing: $1.50/$7.50 per 1M input/output tokens. Cached input tokens receive a 95% discount ($0.075)
Congratulations to the @GoogleDeepMind team on the launch!
@ggerganov v1.0.0 means a breaking change to the public C API should come with a major-version bump. That's a signal for downstream tools, not a promise that the API is frozen forever.
@deepseek_ai Tool fees sit on top of the token bill. $0.75 per 1,000 searches or fetches can add up in a long agent loop. I'd log every tool call from day one.