How Chinese LLMs unlock the secrets of past civilizations through TokenEase's unified API
Archaeology spans millions of years of human history, yet traditional methods remain labor-intensive and slow. A single excavation can take decades to fully publish, while millions of unanalyzed artifacts languish in storage. Chinese LLMs like DeepSeek-V4, GLM-4, and Qwen3 are accelerating discovery by processing satellite imagery, deciphering ancient texts, correlating stratigraphic data, and reconstructing lost environments at unprecedented scales.
Undiscovered sites number in the millions. LLMs can analyze satellite imagery, LiDAR data, historical maps, and environmental variables to identify potential archaeological sites — accelerating survey planning and protecting sites before development destroys them.
import requests
response = requests.post(
"https://tokenease.io/v1/chat/completions",
headers={"Authorization": "Bearer YOUR_TOKENEASE_KEY"},
json={
"model": "deepseek-v4",
"messages": [
{"role": "system", "content": "You are a landscape archaeologist. Analyze environmental and historical data to predict archaeological site locations. Consider: water sources, defensible positions, trade routes, soil types, historical settlement patterns, and modern disturbance. Provide probability maps and survey priorities."},
{"role": "user", "content": """Predict Bronze Age settlement locations in this region:
Geography: River valley, 50x30 km survey area
- Major river: Runs N-S through center, seasonal flooding (May-July)
- Tributaries: 4 minor streams entering from east
- Elevation: 80-340m, valley floor 80-120m, terraces 120-200m
- Soil: Alluvial silt (valley), loess (terraces), rocky uplands (west)
Known sites (excavated):
- Site A: 2100 BCE, 12 hectares, on terrace above floodplain, near river confluence
- Site B: 1950 BCE, 8 hectares, on second terrace, spring nearby
- Site C: 1800 BCE, 25 hectares, at valley narrows (defensive), trade route junction
- Site D: 1750 BCE, 5 hectares, on island in river (religious?)
Environmental data:
- Paleoclimate: 2200-1600 BCE slightly wetter than present
- Sea level: 2m lower than present (estuary 15km south was navigable)
- Copper source: 25km west in uplands
- Obsidian source: 80km east (trade connection confirmed)
Predict 5 most likely undiscovered site locations with rationale."""}
],
"temperature": 0.4,
"max_tokens": 2000
}
)
prediction = response.json()["choices"][0]["message"]["content"]
print(prediction)
# Output: 5 predicted sites with coordinates, probabilities, and rationale:
# - Site E: Terrace above unexamined tributary confluence (85% probability)
# - Site F: Valley narrows south of Site C (70% probability, trade control)
# - Site G: Spring line on east terrace (65% probability, agricultural base)
# - Site H: Estuary fishing station (60% probability, seasonal resource)
# - Site I: Upland pass near copper source (55% probability, extraction camp)Chronological attribution drives archaeological interpretation. LLMs can analyze stylistic features, technological attributes, and compositional data to date artifacts and determine provenance — integrating radiocarbon dates, typological sequences, and stratigraphic relationships.
response = requests.post(
"https://tokenease.io/v1/chat/completions",
headers={"Authorization": "Bearer YOUR_TOKENEASE_KEY"},
json={
"model": "glm-4",
"messages": [
{"role": "system", "content": "You are a ceramic analyst specializing in East Asian pottery. Date and source ceramic artifacts based on: fabric composition, forming technique, surface treatment, decoration, firing conditions, and comparative typology. Provide date ranges with confidence intervals and alternative hypotheses."},
{"role": "user", "content": """Analyze and date this ceramic assemblage:
Context: Tomb M27, burial chamber, sealed context
Sherd 1 (rim):
- Fabric: Fine gray ware, no visible temper
- Form: Carinated bowl, everted rim, ring foot
- Decoration: Incised geometric bands (triangles and parallel lines) below carination
- Surface: Burnished exterior, wiped interior
- Firing: Reduced atmosphere, core dark gray
- Thickness: 4mm wall, 8mm base
- Dimensions: Rim diameter 18cm, height estimated 8cm
Sherd 2 (body):
- Fabric: Same as Sherd 1
- Form: Jar fragment, vertical neck
- Decoration: Applied cord-like strip in spiral pattern
- Surface: Same burnish technique
Comparative data:
- Regional sequence: Late Dawenkou -> Early Longshan -> Middle Longshan -> Late Longshan
- Dawenkou carinated bowls: 4300-2500 BCE, simpler incision, no ring feet
- Early Longshan: 2500-2200 BCE, black pottery, eggshell thin walls, no incision
- Middle Longshan: 2200-2000 BCE, gray ware returns, geometric incision, ring feet appear
- Late Longshan: 2000-1900 BCE, more complex patterns, pedestals, wheel-made
- Western influence (Qijia): 2300-1700 BCE, cord-pattern, different fabric
Provide date range and cultural attribution with confidence."""}
],
"temperature": 0.3,
"max_tokens": 1800
}
)
dating = response.json()["choices"][0]["message"]["content"]
print(dating)
# Output: Attribution: Middle Longshan culture, 2200-2000 BCE (85% confidence)
- Carination + ring foot + geometric incision = Middle Longshan diagnostic
- Absence of eggshell thin walls excludes Early Longshan
- No pedestal or wheel marks excludes Late Longshan
- Cord strip may indicate western contact but not primary attributionUndeciphered scripts — from Linear A to the Indus Valley signs — resist decades of scholarly effort. LLMs can analyze character frequencies, positional patterns, and comparative corpora to propose decipherment hypotheses, assist with damaged inscription restoration, and translate between ancient languages.
response = requests.post(
"https://tokenease.io/v1/chat/completions",
headers={"Authorization": "Bearer YOUR_TOKENEASE_KEY"},
json={
"model": "qwen3-32b",
"messages": [
{"role": "system", "content": "You are an epigrapher specializing in ancient Near Eastern scripts. Reconstruct damaged inscriptions by analyzing: surviving characters, context, formulaic patterns, parallel texts, and linguistic structure. Provide multiple reconstruction hypotheses with confidence levels and philological justification."},
{"role": "user", "content": """Reconstruct this damaged Neo-Assyrian royal inscription (reign of Ashurbanipal, 668-627 BCE):
Medium: Gypsum alabaster wall slab, Nimrud palace
Condition: Upper third lost, middle section damaged, lower third intact
Surviving text (normalized Akkadian, lacunae marked [...]):
[......] ...-ia arad tukulti-ia [...]
[......] ...A SHA KUR [...] sha [...]
[......] ...-ti-ia ina qibit [...]
[......] ...-MU-nu a-na KUR-[...]
[......] ...-ti KUR as+sur ...-ia [...]
[......] ...-ma ina [libbi] [...]
[INTACT] a-na-ku as+sur-ba-ni-ap-li LUGAL GAL LUGAL dan-nu
LUGAL SHA-2 LUGAL KUR as+sur
DUMU as+sur-hu-UD-DUMU-iq-ba LUGAL GAL LUGAL dan-nu
LUGAL SHA-2 LUGAL KUR as+sur
DUMU 10-ERIN.TAH LUGAL GAL LUGAL dan-nu
LUGAL SHA-2 LUGAL KUR as+sur
sa i-na si-qir DINGIR.MES GAL.MES
[......] ...-su-nu [...]
Known parallels: Ashurbanipal's Rassam Cylinder (prism A), lines 1-30 (title formulae)
Expected content: Military campaign against [...], conquest of [...], tribute of [...]
Provide 3 reconstruction hypotheses for the damaged sections."""}
],
"temperature": 0.5,
"max_tokens": 2500
}
)
reconstruction = response.json()["choices"][0]["message"]["content"]
print(reconstruction)
# Output: 3 reconstruction hypotheses with philological justification:
# Hypothesis A (60% confidence): Campaign against Egypt/Nubia (Taharqa/Tanutamani)
# Hypothesis B (25% confidence): Elamite campaign (Teumman/Tammaritu)
# Hypothesis C (15% confidence): Arabian campaign (Qedarites/Arabs)Archaeological stratigraphy — reading layers of soil and debris — reveals chronological sequences and formation processes. LLMs can analyze stratigraphic matrices, correlate depositional events across excavation areas, and distinguish natural from cultural deposits.
response = requests.post(
"https://tokenease.io/v1/chat/completions",
headers={"Authorization": "Bearer YOUR_TOKENEASE_KEY"},
json={
"model": "deepseek-v4",
"messages": [
{"role": "system", "content": "You are a geoarchaeologist specializing in stratigraphic analysis. Interpret excavation layers by analyzing: soil composition, color, texture, inclusions, boundaries, and spatial relationships. Distinguish natural depositional processes from human activities. Identify post-depositional disturbances and dating indicators."},
{"role": "user", "content": """Interpret this stratigraphic sequence from Trench 3 (4x2m, max depth 1.8m):
Layer 1 (0-15cm): Dark brown (10YR 3/3) sandy silt, loose, abundant roots, modern debris (glass, plastic, concrete fragments), sharp lower boundary. Interpretation: Modern topsoil/disturbance.
Layer 2 (15-45cm): Light brown (10YR 5/3) silty clay, compact, occasional gravel, few roots below 30cm, charcoal flecks, sharp wavy lower boundary. Finds: 2 glazed ceramic sherds (19th century), iron nail. Interpretation:?
Layer 3 (45-95cm): Yellowish brown (10YR 5/4) clay with gravel lenses, very compact, no roots, horizontal gravel bands at 55cm and 75cm, charcoal concentrations, gradual lower boundary. Finds: Flint flakes (12), burnt bone fragments, fire-cracked rock. Interpretation:?
Layer 4 (95-140cm): Strong brown (7.5YR 4/6) clay, very compact, no structure, stone alignment (N-S) at 110cm depth, 3 stones visible in section, pit feature (120cm depth, 40cm diameter) filled with dark ash and charcoal, abrupt lower boundary. Finds: Ground stone axe fragment, pottery sherds (undecorated, fabric tempered with quartz), carbonized hazelnut shell. Interpretation:?
Layer 5 (140-180cm): Reddish yellow (7.5YR 6/6) clay with iron concretions, extremely compact, no finds, gradual to natural. Interpretation: Natural subsoil/B-horizon.
Balk disturbance: 20cm wide vertical band in east balk, mixed material from Layers 1-3, probably animal burrow.
Provide interpretation of Layers 2-4, identify formation processes, and suggest dating indicators."""}
],
"temperature": 0.4,
"max_tokens": 2000
}
)
stratigraphy = response.json()["choices"][0]["message"]["content"]
print(stratigraphy)
# Output: Layer 2: Ploughsoil/field disturbance (post-medieval/modern), 19th c. ceramics confirm
- Layer 3: Colluvial/alluvial deposition with prehistoric activity scatter (flint knapping, hunting camp)
- Layer 4: Primary occupation deposit (Neolithic/Bronze Age), stone alignment = wall foundation, pit = storage or refuse, hazelnut shell for C14 datingMillions of archaeological objects await cataloging. LLMs with vision capabilities can analyze artifact photographs, generate structured descriptions, transcribe field notes, and create standardized documentation — dramatically accelerating publication workflows.
response = requests.post(
"https://tokenease.io/v1/chat/completions",
headers={"Authorization": "Bearer YOUR_TOKENEASE_KEY"},
json={
"model": "glm-4v",
"messages": [
{"role": "system", "content": "You are a museum documentation specialist. Generate structured archaeological records from artifact photographs. Include: object type, material, dimensions estimate, manufacturing technique, decoration, condition, and comparative parallels. Use standard museum terminology (CIDOC-CRM compatible)."},
{"role": "user", "content": [
{"type": "text", "text": "Document this artifact from excavation Context 442, Trench 7, Layer 3. Generate complete catalog entry."},
{"type": "image_url", "image_url": {"url": "https://example.com/artifact-t7-l3-442-2026.jpg"}}
]}
],
"temperature": 0.4,
"max_tokens": 1200
}
)
documentation = response.json()["choices"][0]["message"]["content"]
print(documentation)
# Output: Complete catalog entry:
# Object type: Socketed bronze spearhead
# Material: Copper alloy (bronze, likely 90% Cu 10% Sn)
# Dimensions: Estimated 18cm length, 4cm blade width
# Technique: Cast (mold seams visible), socket hammered, blade edges ground sharp
# Decoration: Incised lines on socket, possible herringbone pattern on blade midrib
# Condition: Fragmentary (tip missing, 60% preserved), green patina, stable
# Parallels: Late Bronze Age (Hallstatt A2/B1), Atlantic European typeUnderstanding ancient environments requires synthesizing proxy data — pollen, isotopes, charcoal, faunal remains — across multiple sites and time periods. LLMs can integrate diverse datasets to reconstruct past climates, vegetation, and human-environment interactions.
response = requests.post(
"https://tokenease.io/v1/chat/completions",
headers={"Authorization": "Bearer YOUR_TOKENEASE_KEY"},
json={
"model": "qwen3-32b",
"messages": [
{"role": "system", "content": "You are a paleoclimatologist and environmental archaeologist. Synthesize multi-proxy data to reconstruct past environments and human adaptations. Integrate: pollen records, stable isotopes, sedimentology, faunal remains, and archaeological settlement patterns. Identify climate events, their impacts on human societies, and adaptation strategies."},
{"role": "user", "content": """Reconstruct climate and human environment for this region (8000-3000 BCE):
Proxy data:
- Pollen core (Lake X): 8000-3000 BCE
* 8000-7000 BCE: 80% deciduous oak forest, 10% grasses, warm temperate
* 7000-5500 BCE: 70% oak, 15% grasses, slight opening (Mesolithic clearance?)
* 5500-4500 BCE: 60% oak, 25% grasses, 5% cereal-type pollen (agriculture)
* 4500-3500 BCE: 40% oak, 35% grasses, 15% cereal, beech increase (cooler?)
* 3500-3000 BCE: 30% oak, 40% grasses, 20% cereal, pine increase (cooler, drier)
- Oxygen isotopes (speleothem): Gradual decline from -5.2 to -6.1 (cooler, drier)
- Sedimentation rate: Increased 3x after 4500 BCE (soil erosion from clearance)
- Faunal remains (Site A): 8000-6000 BCE red deer/aurochs dominant; 5500-4000 BCE pig/cattle increase; 4000-3000 BCE sheep dominant, cattle secondary
- Archaeological sites: 8000-5500 BCE 15 open-air sites (seasonal); 5500-4000 BCE 8 settlement sites (permanent, larger); 4000-3000 BCE 12 settlement sites (defensive positions, storage pits)
Interpret: climate trajectory, vegetation change, human adaptation, and identify significant climate events."""}
],
"temperature": 0.4,
"max_tokens": 2000
}
)
paleoclimate = response.json()["choices"][0]["message"]["content"]
print(paleoclimate)
# Output: Climate trajectory: Optimal warm phase 8000-5500 BCE (Atlantic period),
- Gradual deterioration 5500-3000 BCE (cooler, drier, beech/pine expansion),
- No abrupt 8.2ka event detected in this region (buffered by oceanic influence)
- Human adaptation: Seasonal foraging -> permanent agriculture (5500 BCE) -> defensive intensive agriculture (4000 BCE, response to climate stress)
- Soil erosion signal (4500 BCE) indicates intensive land use preceding climate deterioration| Application | Recommended Model | Why |
|---|---|---|
| Site Detection | DeepSeek-V4 | Spatial reasoning, environmental correlation |
| Artifact Dating | GLM-4 | Typological knowledge, comparative analysis |
| Text Decipherment | Qwen3-32B | Pattern recognition, philological reasoning |
| Stratigraphy | DeepSeek-V4 | Multi-factor geological interpretation |
| Documentation | GLM-4V | Vision analysis for artifact description |
| Paleoclimate | Qwen3-32B | Multi-proxy data synthesis |
Access DeepSeek, GLM-4, Qwen3, and vision models through one API.
Start with $1 free credit — no credit card required.