AI models / Alternatives
Claude Fable 5.2 and Mythos 5.2 alternatives, tested on day 7
The other AI models Day 7 has checked, ranked by stamp: Delivered first, then Partly and Broke, with claims still on the record last. Within a stamp, more user reports rank higher.
Claude Fable 5.2 and Mythos 5.2 itself is stamped On record: read its receipt.
- 01Delivered
EmbeddingGemma 2
The open weights, shared multimodal space and smaller sizes check out in independent tests; multilingual parity is shaky outside English, and runtime support arrived late.
- 02Delivered
DeepSeek-V4.1-Flash-1017
Independent benchmarks now confirm the gains over the September build, and it is live and MIT-licensed as promised; it is still verbose and sometimes bends task rules.
- 03Delivered
Claude Opus 5.5
The 20% price cut and 30% speed-up check out, and Epoch and Agent Arena put it at Fable level. It uses more tokens, so cost per task isn't 40% lower.
- 04Delivered
LFM3
Tiny, local and long-context exactly as promised; the 'beats our bigger model' line is still Liquid's own, and it fails at coding.
- 05Delivered
Kimi K3.1 Pro
Weights landed on time and it ranks near the top, but vendor-harness scores shrink in independent runs and it is slow.
- 06Delivered
Muse Spark Open
The weights, licence and specs check out, and it's live on Baseten, Fireworks, Modal, Nebius and OpenRouter. The two-Mac-Studio desktop setup is still untested.
- 07Delivered
GPT-6.1 Sol
Near-Astra performance for less money holds in independent tests, and the prices are as listed. Day-one access was patchy and Ultrafast is top-plan only.
- 08Delivered
Gemini Omni 1.2 Flash
15-second clips and the lower price check out in hands-on tests; edits and the consistency gains are hit and miss.
- 09Delivered
Images 2.6 (GPT-Image-2.6 Flare and Sunburst)
GIFs work and it's live on every plan; the '50% faster' headline depends on which model you pick, and the six-second figure is unmatched.
- 10Delivered
Gemini 4 Carbon
It did top the independent index at launch and opened to everyone two weeks later; the cyber scores are still Google's own runs.
- 11Delivered
Genie 4
The research preview is real and testers see better detail and longer sessions, but worlds still forget what they built, and Google published no numbers.
- 12Partly
Gemini 4 Argon
Live and priced as promised, and the cache discount does work under conditions, but Pro subscribers did not get Antigravity at launch as claimed.
- 13Partly
Mistral Large 4
The weights shipped on time, but the 41% coding headline vanished from the independent board it came from, and 'one of the world's strongest for cyber' is seventh of 25.
- 14Partly
GPT-6.1 Terra
Cheaper and the coding gain is real, but timed tests find it 15 to 25% faster than Sol, not 3x.
- 15On record
Gemini 4 Hydrogen
Day 1: it is live across Google's products, but the benchmark and speed claims are Google's own and independent hands-on reports are only starting.
- 16On record
Meta Muse specialized models (Active Memory and Active Focus)
Rolled out, but every number is Meta's own internal test with no method published, and nobody outside has measured it.
- 17On record
Gemini API price changes (Argon, 3.8 Flash)
Scheduled, not live: Google's page still lists the current rates through 30 Nov. We check the bills from 1 Dec.
- 18On record
GPT-6.2 Astra and Sol
Day 1: the headline wins are OpenAI's own charts, and independent testing so far is mixed, with a top-two ranking overall but no gain on coding agents.
- 19On record
Qwen4-Max
Day 1: the benchmark and cost claims are Alibaba's own, and the model's weights are closed. No independent tests yet.
- 20On record
GPT-6 Cyber
Only approved Daybreak Red customers can use it, so nobody outside can test the claims. The one outside lab in the system card found mixed results.
Ranking uses only the stamp and report count on each receipt. Nothing here is paid placement. How we stamp · All AI models · Submit a launch
Weekly email · Fridays 14:00 CEST
Get the Friday Receipt
Every Friday: the week’s new day 7 receipts, the ones we re-stamped, and any corrections. One email, no sponsors from companies we review.
Prefer not to give an email? Follow the RSS feed or @day7dev on X.