bloopcommitsChhe top models, paanch tests, price aur access, sab ek jagah. Har number ke saath likha hai ki wo company ne khud bataya ya kisi independent tracker ne.
Padhne se pehle
- Ye snapshot reel banate waqt ka hai. Scores aur prices badalte rehte hain.
- "n/a" matlab us source mein number nahi tha. Humne andaza nahi lagaya.
- Vendor-reported = company ne khud bataya. Independent = teesri party ne naapa.
Scores
| Model | Vals Index | DeepSWE | Terminal-Bench | GraphWalks | AA Index |
|---|---|---|---|---|---|
| Gemini 4 Argon | 68.9 | 77.9 | 57.4 | 84.2 | 52.6 |
| GPT-6 Astra | 63.13 | 74.1 | 58.2 | 71.8 | 52.7 |
| GPT-6.1 Sol | 61.15 | n/a | n/a | n/a | 51.8 |
| Claude Opus 5.5 | 66.97 | 74.2 | 66.4 | 66.8 | 57.6 |
| Claude Fable 5.1 | 65.83 | 67.4 | 57.9 | 65.0 | 53.4 |
| Claude Sonnet 5.5 | 67.04 | n/a | n/a | n/a | 56.0 |
Har test kya naapta hai
| Test | Kya naapta hai | Kisne naapa |
|---|---|---|
| Vals Index v2.1 | Kaam-kaaj wale tasks | Independent (Vals AI) |
| DeepSWE v1.1 | Coding | Mixed: Argon ka score Google ne khud nikala |
| Terminal-Bench 4.0 | Terminal ke kaam | Mixed: Argon ka score Google ne khud nikala |
| GraphWalks (256k se 1M) | Lamba context | Google ne saare models ke khud nikale |
| AA Intelligence Index | Overall | Independent (Artificial Analysis) |
Kaun kahan aage
- Vals Index: Argon sabse upar (68.9), phir Sonnet 5.5 (67.04) aur Opus 5.5 (66.97).
- DeepSWE: Argon (77.9), phir Opus 5.5 (74.2) aur Astra (74.1).
- Terminal-Bench: Opus 5.5 saaf aage (66.4). Argon yahan sabse neeche (57.4).
- GraphWalks: Argon aage (84.2).
- AA Index: Opus 5.5 sabse upar (57.6), phir Sonnet 5.5 (56.0). Argon 52.6.
Matlab koi ek model har jagah nahi jeetta. Independent trackers pe tasveer mili-juli hai.
Price (per 1 million tokens, USD)
| Model | Input | Output | Note |
|---|---|---|---|
| Gemini 4 Argon | $2 | $10 | Shuruaati price. Baad mein $4 / $20. |
| GPT-6 Astra | $10 | $50 | Secondary source se. |
| GPT-6.1 Sol | $2 | $10 | Secondary source se. |
| Claude Opus 5.5 | $4 | $20 | Official list price. |
| Claude Fable 5.1 | $10 | $50 | Official list price. |
| Claude Sonnet 5.5 | $2 | $10 | Official list price. |
Per-token price aur per-task kharcha alag cheezein hain. Jo model ek kaam mein zyada tokens likhta hai, wo sasta dikh ke bhi mehnga pad sakta hai.
Aaj kaun use kar sakta hai
| Model | Access |
|---|---|
| Gemini 4 Argon | Limited: sirf Fairwind Program ke cyber partners. |
| GPT-6 Astra | ChatGPT ke paid plans aur API. Free tier pe nahi. |
| GPT-6.1 Sol | Paid plans aur API. Free tier pe nahi. |
| Claude Opus 5.5 | Claude ke paid plans aur API. |
| Claude Fable 5.1 | Claude ke paid plans aur API. |
| Claude Sonnet 5.5 | Anthropic ke hisaab se koi bhi Claude.ai pe isse chat kar sakta hai. |
Apne liye model kaise chunein
- Pehle dekho tumhe access kiska hai. Jo mil hi nahi sakta, uska score kaam ka nahi.
- Apna asli task 2 models pe chalao: wahi prompt, wahi input.
- Nateeja dekho: sahi hai ya nahi, kitna theek karna pada.
- Phir price dekho, per-task ke hisaab se.
Sources
Vals Indexwww.vals.ai/benchmarks/vals_index
Artificial Analysisartificialanalysis.ai/models/gemini-4-argon
Google blog: Gemini 4 Argonblog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
Anthropic: Claude Opus 5.5www.anthropic.com/news/claude-opus-5-5
Anthropic: Claude Sonnetwww.anthropic.com/claude/sonnet