
GPT-6.2 Astra & Sol Are Out! Slightly Underwhelming, Though...
GPT-6.2 Astra hits 62 on the AA index at $8/$40 to retake #2 from Gemini 4 Carbon, while Fable 5.2 holds #1 at 65. Speed, cost, Codex reactions, and the two-month delay.
Models, robotics, on-device AI, and practical applications in one place.

GPT-6.2 Astra hits 62 on the AA index at $8/$40 to retake #2 from Gemini 4 Carbon, while Fable 5.2 holds #1 at 65. Speed, cost, Codex reactions, and the two-month delay.

Gemini 4 Hydrogen hits AI Pro: AA 54 at xhigh, 188 tok/s, $1.50/$7.50 per 1M tokens. Carbon is still off-limits on Pro, so can cheap, fast Hydrogen be the relief pitcher?

It tops the AA index at 65, yet it is only 5.2. Why I think 5.5 is being held back, the METR and S-1 evidence, and the tests where rivals still win.

Artificial Analysis Index v5 hides 60% of its score behind private tests, so labs can no longer tune to the questions. Why that matters, what Kaggle does, and what it can't fix.

Gemini 4 Carbon tops the Artificial Analysis index at 59, nine points over Argon at the same price, yet Vals and Arena call it close to a tie. Nobody can use it yet.

GPT-6.1 Terra costs $1/$5 per 1M tokens, half of Sol. From GPT-3 at $60 to now, the price of the same benchmark score fell about 860x in three years.

Claude Duet is out for US Pro and Max on a waitlist: a morning brief and check-ins every few hours. Next to dots, Muse and Grok Bot, it asks first at every sensitive step.

The fastest Ender Dragon kill took 8m 43s, but a scripted route did most of the work. Under the same screen-and-keyboard rules, no model even found iron.

Gemini 4 Argon is generally available and hallucinates far less, but each question took about 5% of my AI Pro meter, and Antigravity shut out Pro at launch.

Muse Wonder Full scores 56 on the Artificial Analysis index, between Claude Opus 5.5 and GPT-6 Astra. What the name, prices, Meta's own table and the Muse app show.

Haiku 5.5 leads Luna in published Max-effort tests, but long prompts cost more; I plan to move my main subscription to Claude while watching for an OpenAI response.

Gemini 4 Argon promises frontier performance, but access and a public date remain unclear; an Ultra-first plan is a hard sell before a roughly $100 monthly upgrade.

Voxel and animation demos fuel Fable 5.5 rumors, but the catalog still lists 5.1; GameCube and Superman clips add variety without confirming the model's identity.

Argon has strong scores and appealing launch prices, but past Antigravity frustrations make a one-month trial more convincing than annual payment; wider access has no set date.

Dots is the persistent AI assistant I want to try for stock research, work management and this blog. What OpenAI launched, and how I would test it.

DevDay felt substantial because models, agents, developer tools and collaboration improved together: all 25 announcements, their rollout limits, and why Sol stands out to me.

At High, Sonnet 5.5 gave me good value on everyday work, but at Max it used about 60% more tokens than Opus 5.5 and cost $7.60 per task against $5.98

Claude Code and Codex tie at 57.9% on Terminal-Bench 4.0, but Codex spends half as much per attempt. Antigravity is free with Claude 4.6, Cursor favors Grok, Muse Code costs $5

Behind OpenAI and Anthropic, Meta, Grok and five Chinese labs score 39–48. Only Xiaomi and DeepSeek are cheap per task, and Anthropic says all five Chinese labs distilled Claude

Grok 5 missed Musk's Q1 promise, and Terafab and orbital compute are 2028 projects at the earliest. Meanwhile Anthropic and Google rent Colossus for $2.17B a month

Sonnet 5.5 follows the pacing essay without breaking it, the gated Mythos is the same model as public Fable, and the leaked price is just Sonnet 5's current rate

Astra Minor is one vanished line in an Azure file, Sol slid from flagship to middle tier, and Opus 5.5 now beats Astra on most of Anthropic's table at 40% of the price

Google says Gemini 4 is in post-training and due well before year-end. Bettors lean October, the leaked scores and $2.25 price have no source, and Antigravity gets it first