The right AI for your work is a matter of record.
Four labs, dozens of models, one that actually fits how you work. We keep the benchmarks current, read the fine print, and match you to it in three minutes.
Today’s standings, asterisks included.
Independent numbers where they exist, vendor numbers labeled where they don’t. This table changes when the facts do.
| Model | Intelligence¹ | Output $/1M | The verdict |
|---|---|---|---|
| Claude Opus 5.5 | 58#1 | 20.00 | Still first, now by two points, at $20 out; AA's measured cost per task is $5.98, within 9% of Sonnet 5.5's. |
| Claude Sonnet 5.5NEW | 56#2 | 10.00 | Half of Opus 5.5's price per token and the most verbose model AA has measured; at max effort the bill is $5.46 per task. |
| Muse Wonder FullNEW | 56#2 | -⁴ | Ties Sonnet 5.5 and leads AA's legal-agent board, but Meta sells it only inside a $50 or $200 Muse Code subscription. |
| Claude Fable 5.1 | 53#4 | 50.00 | Tied fourth at 2.5x Opus 5.5's price, with the highest cost per task of any flagship AA charts, $7.63. |
| GPT-6 Astra | 53#4 | 50.00³ | Tied fourth at $50 out, bills the whole request at $75 past 272K input, and its 6.1 successor was never released. |
| Gemini 4 ArgonNEW | 53#4 | 10.00² | Google's first Gemini 4 ties Astra at 60% of its cost per task with the lowest hallucination rate on the board; the price doubles December 1. |
| GPT-6.1 SolNEW | 52#7 | 10.00³ | One point under Astra at $0.72 per AA task against $3.26, and worse than Astra on unwanted persistence in its own system card. |
| Muse WonderNEW | 52#7 | 20.00 | Opus 5.5's list price for six fewer points; the API model is a preview, and Meta's better model is not on the API. |
| Grok 4.7 | 46#9 | 6.00⁷ | Still $2/$6, still verbose: AA's finished task costs $3.74, more than GPT-6 Astra's $3.26, despite the lowest per-token price here. |
| Claude Haiku 5.5NEW | 43#10 | 0.50⁶ | GPT-6 Luna's list price for five more points, but about 3x Luna's tokens per task and a 5x surcharge past 100K tokens. |
| Gemini 3.8 Flash | 41#11 | 3.75⁵ | Still the fastest workhorse here, now outscored by Haiku 5.5 at a lower list price; the sale ends December 31. |
| GPT-6 Luna | 38#12 | 0.50 | Seven cents per AA task and a 77% hallucination rate; fine for extraction and the Decisions API, not for facts. |
¹ Artificial Analysis Intelligence Index v4.3.2, measured independently, read October 16, 2026, at each model’s highest scored effort (max for the Claude, GPT-6, and Muse models, xhigh for Grok 4.7, high for Gemini 4 Argon and Gemini 3.8 Flash); ranks are within this table. Same index version as our September table, so those scores are comparable; the August table was on a retired scale. Claude Opus 5 (51) and GPT-6 Sol (48) stay on sale and are omitted here. ² Introductory rate through Nov 30, 2026; doubles to $20 on Dec 1. ³ Requests over 272K input tokens bill the whole request at a higher rate ($75 for Astra, $15 for GPT-6.1 Sol). ⁴ No API; sold only inside Muse Code at $50 or $200 a month. ⁵ Introductory rate through Dec 31, 2026; doubles to $7.50 on Jan 1, 2027. ⁶ Prompts over 100K tokens bill the whole request at $2.50. ⁷ Prompts at or above 200K tokens bill the whole request at $12.
Prices are list API rates per 1M output tokens. Dashes mean we won’t print a number we haven’t checked.
Field notes, filed under oath.
Every model launch, separated into what’s verified and what’s marketing. No press-release stenography.
gpt-6-astra-law Is on the Pricing Page at Plain Astra's Rate, Labeled Promotional Through December 31. FedRAMP Endpoints Get a 10% Uplift From November 1.
OpenAI's API pricing page lists gpt-6-astra-law under legal model configurations at $10 input, $50 output, and $1 cached per million tokens, with a note that the promotional price holds at least through December 31, 2026.
→Nº 126 · OCT 30Clover-35B-A3B Claims 73.6 on SWE-bench Verified. Every Number Is the Vendor's, and the Decision Model Is Not Out.
DiffuSpace released Clover-35B-A3B-Preview on October 30: a diffusion mixture-of-experts model derived from Qwen3.6-35B-A3B, 35 billion total parameters with about 3 billion active, a 128K context, and an Apache license, with weights on Hugging Face.
→Nº 125 · OCT 30OpenAI, Google, Microsoft, and Anthropic Will Honor One robots.txt No-AI-Training Token. Only Microsoft Has Given a Date: February 1, 2027.
On October 30 the four labs said they will honor a shared robots.txt preference from the IETF AIPREF draft: a User-Agent of __usage_ai-training with a Disallow line.
→Nº 124 · OCT 30Claude Max 20x Is Now Ultra at the Same $200. The 20x Was Only Ever the Five-Hour Cap; Weekly Stays at 10x Pro.
On October 30 Anthropic renamed Max 5x to Max at $100 and Max 20x to Ultra at $200, with no price change.
→Written by T.W. Ghost · corrections welcome, receipts required
Three tracks. Pick your altitude.
Ordered by how much you already know, not by how much we can sell you.
PRO TRACKS · 38 industry courses, from AI for Marketing to AI for Day Traders.