Skip to content
AI Atlas
Ecosystem pulse

The last 7 days in AI

Deterministic counters over events that occurred in the window — historical backfill excluded, so a re-crawl of old pages never inflates them. Hover a label for how each one is counted.

5 Sept 2026 → 12 Sept 2026 07:07 UTC

Window

New modelscanonical models with a NEW_MODEL event that occurred in the window (not back-filled)

102

New open-weight modelssubset of new_models with openness open-weights / open-source

84

New artifactsartifact entities (checkpoints, quantisations, conversions) first seen in the window

174

New paperspapers with a NEW_PAPER event that occurred in the window

418

Provider listingsPROVIDER_LISTED events in the window

15

DelistingsPROVIDER_DELISTED events in the window

0

Price changesPRICE_CHANGED events in the window; % change = output price (else input) new vs old from the event's old/new values; median over events with both values

20

New ≥ 1M-context modelsnew_models whose context_length is at least 1 000 000 tokens

0

New benchmark leadersbenchmarks whose primary-group leader (computed from results observed by each date) changed over the window

23

Documents changedDOCUMENT_CHANGED events in the window

235

Sources observeddistinct sources with at least one snapshot taken in the window

26

Events totalall non-backfill events (excluding source-document changes) that occurred in the window

1,432

All counters use is_backfill = false and occurred_at inside the window. Sparklines show the daily stats history where a series exists (2 days recorded so far).

MethodDeterministic counters over events that OCCURRED in the window and are not back-filled history; each counter carries its own definition. /methodology

Benchmark leadership

New benchmark leaders · 23

benchmarks whose primary-group leader (computed from results observed by each date) changed over the window

All leaderboards →
New benchmark leaders
BenchmarkCurrent leaderScorePrevious leaderGroupTrust
SWE-bench VerifiedClaude Opus 4.5Anthropic76.8 resolvedfirst leader recordedresolved · board=Verified · system=mini-SWE-agent · n=42official benchmark
Aider polyglotgpt-5OpenAI86.7 pass_rate_2first leader recordedpass_rate_2 · n=61official benchmark
LiveBenchClaude Fable 5.1Anthropic83.41 global_averagefirst leader recordedglobal_average · n=57official benchmark
Humanity's Last ExamClaude Fable 5.1Anthropic59.13 accuracyfirst leader recordedaccuracy · evaluator=Artificial Analysis · n=453independent evaluator
Terminal-Benchgpt-5.6-solOpenAI65.91 accuracyfirst leader recordedaccuracy · variant=hard · evaluator=Artificial Analysis · n=315independent evaluator
Artificial Analysis Intelligence IndexClaude Fable 5.1Anthropic53.37 indexfirst leader recordedindex · n=477independent evaluator
SWE-bench Liteclaude-3-opusAnthropic4.33 resolvedfirst leader recordedresolved · board=Lite · system=RAG baseline · n=6official benchmark
SWE-bench (full test split)claude-3-opusAnthropic3.79 resolvedfirst leader recordedresolved · board=Test · system=RAG baseline · n=6official benchmark
SWE-bench Multimodalo3OpenAI35.98 resolvedfirst leader recordedresolved · board=Multimodal · system=GUIRepair · n=4official benchmark
SWE-bench Multilingualgemini-3-flashGoogle72.7 resolvedfirst leader recordedresolved · board=Multilingual · system=mini-SWE-agent · n=13official benchmark
MMMU-Progpt-6-astraOpenAI86.88 accuracyfirst leader recordedaccuracy · evaluator=Artificial Analysis · n=156independent evaluator
SciCodeClaude Fable 5.1Anthropic63.08 accuracyfirst leader recordedaccuracy · evaluator=Artificial Analysis · n=126independent evaluator
IFBenchGrok 4.3xAI83.33 accuracyfirst leader recordedaccuracy · evaluator=Artificial Analysis · n=333independent evaluator
τ²-benchZ.ai GLM 5.2Z.ai (Zhipu AI)99.12 pass^1first leader recordedpass^1 · evaluator=Artificial Analysis · n=323independent evaluator
GPQA Diamondgpt-6-astraOpenAI96.26 accuracyfirst leader recordedaccuracy · variant=GPQA Diamond · evaluator=Artificial Analysis · n=459independent evaluator
Aider polyglot — well-formed responsesCodestral 25.01Mistral AI100 percent_cases_well_formedfirst leader recordedpercent_cases_well_formed · n=61official benchmark
LiveBench Reasoninggpt-6-astraOpenAI92.65 average scorefirst leader recordedaverage score · variant=Reasoning · n=57official benchmark
LiveBench CodingClaude Fable 5.1Anthropic86.38 average scorefirst leader recordedaverage score · variant=Coding · n=57official benchmark
LiveBench Agentic CodingDeepSeek-V4.1-FlashDeepSeek77.27 average scorefirst leader recordedaverage score · variant=Agentic Coding · n=57official benchmark
LiveBench MathematicsClaude Fable 5.1Anthropic97.01 average scorefirst leader recordedaverage score · variant=Mathematics · n=57official benchmark
LiveBench Data Analysisgpt-6-astraOpenAI82.97 average scorefirst leader recordedaverage score · variant=Data Analysis · n=57official benchmark
LiveBench LanguageClaude Fable 5Anthropic90.68 average scorefirst leader recordedaverage score · variant=Language · n=57official benchmark
LiveBench Instruction FollowingGemini 3.8 FlashGoogle81.41 average scorefirst leader recordedaverage score · variant=IF · n=57official benchmark

Prices

Price changes · 20 · median -0.41 %

PRICE_CHANGED events in the window; % change = output price (else input) new vs old from the event's old/new values; median over events with both values

Price terminal →

The pulse returns the count only; the rows below are PRICE_CHANGED events since 5 Sept 2026 from the change feed (same window, same backfill rule).

Recent price changes
ModelProviderInput / 1MOutput / 1MChangeObservedSource
DeepSeek V4 Flash 0423DeepSeekDeepSeek API$0.067$0.067↓$0.134$0.134↓in −0.4%out −0.4%1 h agoopenrouter.ai
DeepSeek V4 Pro 0423DeepSeekDeepSeek API$0.829$0.819↓$1.66$1.64↓in −1.2%out −1.2%1 h agoopenrouter.ai
DeepSeek V4 Flash (0731)DeepSeekDeepSeek API$0.065$0.04↓$0.18$0.08↓in −38%out −56%1 h agoopenrouter.ai
DeepSeek V4 Flash 0423DeepSeekDeepSeek API$0.067$0.067↑$0.134$0.134↑in +0.4%out +0.4%2 h agoopenrouter.ai
DeepSeek V4 Flash 0423DeepSeekDeepSeek API$0.067$0.067↓$0.135$0.134↓in −0.8%out −0.8%3 h agoopenrouter.ai
DeepSeek V4 Pro 0423DeepSeekDeepSeek API$0.85$0.829↓$1.7$1.66↓in −2.4%out −2.4%3 h agoopenrouter.ai
Llama 3.1 70B InstructMeta AIOpenRouter$0.40$0.72↑$0.40$0.72↑in +80%out +80%6 h agoopenrouter.ai
Qwen3 235B A22B Instruct 2507QwenOpenRouter$0.22$0.087↓$0.88$0.35↓in −60%out −60%6 h agoopenrouter.ai
MiniMax M2.5MiniMaxMiniMax API$0.27$0.30↑$1.08$1.2↑in +11%out +11%6 h agoopenrouter.ai
DeepSeek V4 Flash 0423DeepSeekDeepSeek API$0.068$0.067↓$0.135$0.135↓in −0.4%out −0.4%6 h agoopenrouter.ai
DeepSeek V4 Pro 0423DeepSeekDeepSeek API$0.84$0.85↑$1.68$1.7↑in +1.2%out +1.2%6 h agoopenrouter.ai
Kimi K3Moonshot AIMoonshot AI Platform$1.54$2.3↑$7.7$11.55↑in +50%out +50%6 h agoopenrouter.ai
DeepSeek-V4-Pro-0813DeepSeekDeepSeek API$0.579$0.578↓$1.74$1.73↓in −0.2%out −0.2%6 h agoopenrouter.ai
Qwen3.8 27BQwenOpenRouter$0.42$0.214↓$3$2.55↓in −49%out −15%6 h agoopenrouter.ai
DeepSeek V4 Flash 0423DeepSeekDeepSeek API$0.085$0.068↓$0.169$0.135↓in −20%out −20%7 h agoopenrouter.ai
DeepSeek V4 Pro 0423DeepSeekDeepSeek API$0.948$0.84↓$1.9$1.68↓in −11%out −11%7 h agoopenrouter.ai
Hy3TencentOpenRouter$0.083$0.132↑$0.33$0.528↑in +60%out +60%7 h agoopenrouter.ai
Kimi K3Moonshot AIMoonshot AI Platform$1.8$1.54↓$9.01$7.7↓in −15%out −15%7 h agoopenrouter.ai
DeepSeek V4 Flash 0423DeepSeekDeepSeek API$0.085$0.085↓$0.17$0.169↓in −0.3%out −0.3%9 h agoopenrouter.ai
DeepSeek V4 Pro 0423DeepSeekDeepSeek API$0.86$0.948↑$1.72$1.9↑in +10%out +10%9 h agoopenrouter.ai

Movers are PRICE_CHANGED events emitted when a provider's published price for a model differs from the previous observation. Green = cheaper, red = dearer; % is relative to the previous published price.

Long context

New ≥ 1M-context models · 0

new_models whose context_length is at least 1 000 000 tokens

No new model with a ≥ 1M-token context in the last 7 days

For reference, the most recently released models with a sourced context of at least 1M tokens (any date):

Reading the pulse

Definitions

New models 102
canonical models with a NEW_MODEL event that occurred in the window (not back-filled)
New open-weight models 84
subset of new_models with openness open-weights / open-source
New artifacts 174
artifact entities (checkpoints, quantisations, conversions) first seen in the window
New papers 418
papers with a NEW_PAPER event that occurred in the window
Provider listings 15
PROVIDER_LISTED events in the window
Delistings 0
PROVIDER_DELISTED events in the window
Price changes 20
PRICE_CHANGED events in the window; % change = output price (else input) new vs old from the event's old/new values; median over events with both values
New ≥ 1M-context models 0
new_models whose context_length is at least 1 000 000 tokens
New benchmark leaders 23
benchmarks whose primary-group leader (computed from results observed by each date) changed over the window
Documents changed 235
DOCUMENT_CHANGED events in the window
Sources observed 26
distinct sources with at least one snapshot taken in the window
Events total 1,432
all non-backfill events (excluding source-document changes) that occurred in the window

Source: GET /pulse?days=7. Nothing on this page is a projection.