Machine-readable
The data
The canonical comparison records. Each row carries source details or claim IDs that resolve in the edition's evidence register. Sorting is a reading aid; it does not establish comparability.
Use this in your own work. These tables and the report are
CC BY 4.0 — quote them, chart them, or build on them in
your articles and reports (commercial use is fine). Just credit State of Agentic Coding with a link
back. Grab the raw CSVs below, or cite the edition: State of Agentic Coding, edition
2026-09/openai-gpt-6, https://github.com/lbotinelly/state-of-agentic-coding.
Evidence records
Claims ledger · Source records · Research log
Confidence, supporting passages, conflicting findings, retrieval dates, and research limitations remain attached to the records.
Model and harness setups
| Id | Model | Harness | Harness Version | Configuration | Task Context | Availability | Access Route | Strengths | Limitations | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|
| astra | gpt-6-astra | Codex CLI/IDE local | unknown | Start high; escalate effort only after checking failure | Complex implementation and debugging | API paid tiers; client access depends on account | OpenAI direct API or eligible Codex plan | Steerable long-running candidate | No model-only superiority established; costly cache/long requests | CR01;CR02;CH17;CH20 |
| fable51 | claude-fable-5-1 | Claude Code | unknown | High default; preserve prefixes and verify fallback route | Hard refactoring and long sessions | GA; Pro/Max/Team/Enterprise and listed clouds | Anthropic native; intermediary terms separate | Cache-heavy escalation candidate | Retention eligibility; API breaking changes; human review needed | CR03;CR04;CR05;CR06;CH05;CH06 |
| opus5 | claude-opus-5 | Claude Code | unknown | High; bounded task and tests | Everyday implementation | Available since July 24 | Claude or native API | Lower list rates than Fable; vendor recommended starting model | Vendor guidance; no measured repo-specific preference | CR07;CR16 |
| sol | gpt-5.6-sol | Codex | unknown | Medium starting effort; bounded task | Everyday implementation | Listed active model; account access varies | OpenAI native API or eligible Codex plan | Lower current token rates | Promotion; no accepted-task savings established | CR13;CH17 |
| gemini38 | gemini-3.8-flash | Antigravity managed agent | unknown | Medium default; allow billed reasoning | Greenfield/UI and bounded budget evaluation | GA September 2 | Google managed agent; API separately billed | Native managed route; managed-agent economics unmeasured | May spend more tokens; no independent advantage established here | CR08;CR09 |
| deepseek4 | deepseek-v4-pro | Codex | unknown | Vendor Responses configuration; high starting effort | Budget workflow compatibility evaluation | GA hosted Pro0813; Flash0731 separately | DeepSeek API | Documented Responses integration | Not tested; Responses protocol has unsupported fields and tools | CR11;CR12;CR18 |
| glm53 | glm-5.3 | OpenCode | unknown | Reasoning enabled; explicitly select effort; retain task tests | Budget-conscious implementation and debugging with human review | All Coding Plan tiers; API account restrictions documented | Z.ai Coding Plan or separate API | Low posted token rates; documented Coding Plan support | Task acceptance cost unmeasured; aliases reroute; account protocol restrictions; Coding Plan privacy mapping unresolved | CGL01;CGL04;CGL06;CGL20;CGLP03 |
| glm53eval | glm-5.3 | Claude Code | 2.1.207 | max effort; 400K context; 128K output; temperature=1; top_p=1; Tool Search disabled | Vendor Terminal-Bench 3.0 evaluation | Vendor reported result | Vendor evaluation; inference route unknown | Detailed vendor evaluation conditions | No independent reproduction or matched deployment measurements | CGL02 |
| glm53flash | glm-5.3-flash | OpenCode | unknown | Explicit effort; max default; multimodal model | Low-price, bounded implementation tasks to pilot | All Coding Plan tiers; open weights available | Z.ai Coding Plan or separate API | Lower durable list rates; multimodal input; MIT weights | Quality and accepted-task savings unverified; not a small local model | CGL05;CGL06;CGL08 |
| local8 | Qwen/Qwen3.5-4B | Qwen-Agent + vLLM/SGLang | unknown | Proposed 4-bit weights; KV precision and context to measure | 8 GB dedicated VRAM capacity-screening pilot | Released weights or documented service through cutoff | Self-hosted | Small Apache 2.0 candidate with documented local agent route | Quant format, runtime memory and task reliability not tested | CGL17;CGL18;CGL21 |
| local16 | Qwen/Qwen3.5-9B | Qwen-Agent + vLLM/SGLang | unknown | Proposed 8-bit weights; KV precision and context to measure | 16 GB dedicated VRAM capacity-screening pilot | Released weights or documented service through cutoff | Self-hosted | Larger small-model candidate with local MCP route | No measured agentic reliability or latency | CGL17;CGL18;CGL21 |
| local24 | Qwen/Qwen3.8-27B | vLLM/SGLang endpoint + chosen agent | unknown | Proposed 4-bit weights; preserve reasoning per model docs; client verification required | 24 GB dedicated VRAM local pilot | Released weights or documented service through cutoff | Self-hosted | New dense model; Apache 2.0 designation; vendor coding improvement | Vendor benchmark does not validate this 4-bit setup; client interoperability untested | CGL09;CGL10;CGL18;CGL21 |
| localbig | Qwen/Qwen3.8-27B | Selected local serving stack and agent | unknown | Proposed 8-bit weights; engine/hardware support to verify | 64 GB unified-memory capacity-screening pilot | Released weights or documented service through cutoff | Self-hosted | More weight precision headroom than the 24 GB tier | Unified memory shared with OS; neither engine compatibility nor latency measured | CGL09;CGL18;CGL21 |
| qwen38eval | Qwen/Qwen3.8-27B | Terminus | unknown | Vendor benchmark; detailed configuration unknown | Vendor Terminal-Bench 2.1 result | Vendor reported result | Vendor evaluation; route unknown | Reported score of 73.0 | No 4-bit or commodity GPU inference established | CGL10 |
| kimik3 | kimi-k3 | Kimi Code CLI | unknown | Preserve full assistant messages; effort low/high/max; max default | Long-session engineering candidate | Documented API and released weights | Kimi API; weights also available | Native multimodality and 1M context documented | Vendor capability claims; custom license; huge self-hosting memory | CGL13;CGL14;CGL18 |
| minimaxm3 | MiniMax-M3 | Selected compatible agent | unknown | Choose enabled/adaptive/disabled thinking; integration test required | Multimodal coding candidate | Released weights or documented service through cutoff | MiniMax API; weights also available | Documented 1M context and serving routes | Custom license; no measured task economics or client validation in this review | CGL15;CGL16 |
| qwen38next | Qwen/Qwen3.8-Flash-Next | Selected compatible agent | unknown | Experimental preview; version/quantization and parser verification required | Larger-server experiment | Open-weight experimental architecture preview | Self-hosted | Sparse architecture with offloadable embeddings | Custom AI Work Assistant/MaaS licensing; 6B active parameters do not determine total weight memory | CGL11;CGL12 |
| TO1 | GPT-6 Astra | Codex | unknown | Max effort; Harbor evaluation using Terminal-Bench 4.0.0; remaining settings unknown | 66 terminal-work tasks in Terminal-Bench 4.0.0 | Observed benchmark submission; current product access documented separately | unknown | Published terminal-task resolution result | Exact harness version, fallback routing, complete budget and evaluation date unknown; descriptive result only | CO04;CO03 |
| TO2 | Claude Fable 5.1 | Claude Code | unknown | Max effort; Harbor evaluation using Terminal-Bench 4.0.0; remaining settings unknown | 66 terminal-work tasks in Terminal-Bench 4.0.0 | Observed benchmark submission; current product access documented separately | unknown | Published terminal-task resolution result | Exact harness version, fallback routing, complete budget and evaluation date unknown; descriptive result only | CO05;CO03 |
| TO3 | GLM-5.3 | Claude Code | unknown | Max effort; Harbor evaluation using Terminal-Bench 4.0.0; remaining settings unknown | 66 terminal-work tasks in Terminal-Bench 4.0.0 | Observed benchmark submission; current product access documented separately | unknown | Published terminal-task resolution result | Exact harness version, fallback routing, complete budget and evaluation date unknown; descriptive result only | CO06;CO03 |
| TO4 | Claude Fable 5.1 | Claude Code | unknown | Endor Labs Agent Security League evaluation; exact inference settings unknown | Python tasks derived from CVE patches | Benchmark row dated 2026-09-04 | unknown | Functional and security-test outcomes reported separately | Exact September run settings, adjusted denominator and uncertainty unknown | CO07 |
| copilot-team | gpt-6-astra | GitHub Copilot app/CLI | unknown | Business/Enterprise policy; exclusions and user budgets; human review | Existing GitHub team with approved hosted-data route | GA rollout gradual; admin model enablement | Copilot Business/Enterprise | Central policy and credit controls; documented Astra route | Provider ZDR does not establish GitHub service retention; task advantage unmeasured | CH02;CH04;CH07;CH12;CH21 |
Benchmark observations
Compare only within a declared group with a compatible evaluation design. Rows without a group are descriptive; unknown is not zero. Inspect all conditions below before interpreting scores.
| Id | Setup Id | Benchmark | Benchmark Version | Split | Evaluator | Evidence Type | Comparison Group | Comparison Basis | Comparison Notes | Tools | Resource Budget | Time Limit | Token Budget | Attempts | Metric | Score | Unit | Sample Size | Uncertainty | Contamination Notes | Fallback Routing | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| glm53tb3 | glm53eval | Terminal-Bench | 3.0 | unknown | Z.ai | vendor-experiment | none | none | Descriptive only; sample/uncertainty and cross-model conditions incomplete. | Claude Code tools; Tool Search disabled | Official task containers; 600 agent turns | 10 hours per rollout | 400K context; 128K maximum output | 3 | avg@3 official verifier score | 28.3 | reported score (0–100) | unknown | unknown | No contamination audit established in this review. | unknown | CGL02 |
| qwen38tb21 | qwen38eval | Terminal-Bench | 2.1 | unknown | Qwen | vendor-experiment | none | none | Descriptive only; sample/uncertainty and cross-model conditions incomplete. | Terminus; tool details unknown | unknown | unknown | unknown | unknown | Vendor reported score | 73.0 | reported score (0–100) | unknown | unknown | No contamination audit established in this review. | unknown | CGL10 |
| BO1 | TO1 | Terminal-Bench | 4.0.0 | Published 66-task dataset; all-task evaluation | Terminal-Bench/Harbor; contributor presentation by Snorkel AI | independent-experiment | none | none | Descriptive system observation. Missing harness versions, attempts, exact resource budgets and error-bar definition preclude a ranked group. | Terminal execution; exact enabled tool set unknown | Per-task CPU/memory recalibrated; exact values unknown | 8 hours per task | unknown | unknown | Resolution rate | 58.2 | percent | 66 | Reported ±2.8 percentage-point margin reflecting run-to-run variance; interval level unknown | Public dataset and canary request; actual training overlap unknown | unknown | CO04;CO03 |
| BO2 | TO2 | Terminal-Bench | 4.0.0 | Published 66-task dataset; all-task evaluation | Terminal-Bench/Harbor; contributor presentation by Snorkel AI | independent-experiment | none | none | Descriptive system observation. Missing harness versions, attempts, exact resource budgets and error-bar definition preclude a ranked group. | Terminal execution; exact enabled tool set unknown | Per-task CPU/memory recalibrated; exact values unknown | 8 hours per task | unknown | unknown | Resolution rate | 57.9 | percent | 66 | Reported ±3.8 percentage-point margin reflecting run-to-run variance; interval level unknown | Public dataset and canary request; actual training overlap unknown | unknown | CO05;CO03 |
| BO3 | TO3 | Terminal-Bench | 4.0.0 | Published 66-task dataset; all-task evaluation | Terminal-Bench/Harbor; contributor presentation by Snorkel AI | independent-experiment | none | none | Descriptive system observation. Missing harness versions, attempts, exact resource budgets and error-bar definition preclude a ranked group. | Terminal execution; exact enabled tool set unknown | Per-task CPU/memory recalibrated; exact values unknown | 8 hours per task | unknown | unknown | Resolution rate | 41.8 | percent | 66 | Reported ±3.2 percentage-point margin reflecting run-to-run variance; interval level unknown | Public dataset and canary request; actual training overlap unknown | unknown | CO06;CO03 |
| BO4 | TO4 | Agent Security League / SusVibes | unknown | Python tasks derived from CVE fixes; 200 original and 179 feasible cases in the method; exact row denominator unknown | Endor Labs | independent-experiment | none | none | Security vendor's experiment on third-party agents. Descriptive; exact September run metadata and uncertainty unavailable. | unknown | unknown | unknown | unknown | unknown | FuncPass | 87.2 | percent | unknown | unknown | Workspace sanitization, prompt hardening, cheating detection and score adjustment; public fixes remain a leakage risk | unknown | CO07 |
| BO5 | TO4 | Agent Security League / SusVibes | unknown | Python tasks derived from CVE fixes; 200 original and 179 feasible cases in the method; exact row denominator unknown | Endor Labs | independent-experiment | none | none | Security vendor's experiment on third-party agents. Descriptive; exact September run metadata and uncertainty unavailable. | unknown | unknown | unknown | unknown | unknown | SecPass | 37.4 | percent | unknown | unknown | Workspace sanitization, prompt hardening, cheating detection and score adjustment; public fixes remain a leakage risk | unknown | CO07 |
Conditional recommendations
| Id | Audience | Scenario | Prerequisites | Setup Id | Pick | Alternative | Tradeoffs | Confidence | Confidence Reason | Would Change If | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|---|
| everyday | Individual | Daily implementation | Tests; review | sol | Keep Sol/Codex; reassess | Opus 5/Claude Code | Promotional rates; task costs unmeasured | medium | Documented rates; limited experience | Another setup reduces accepted-task cost | CR13;CR07;CR16;CH17;CO12 |
| hard | Experienced developer | Hard debugging/refactoring | Repro case; approved data route | astra | Trial Astra/Codex local | Fable 5.1/Claude Code | No clear leader; Fable retention/migration limits | medium | Relevant system evidence; incomplete conditions | Repo trials favor alternative | CR01;CR02;CR03;CR04;CR05;CH17;CO04;CO05 |
| budget | Cost-sensitive developer | Bounded volume | Tests; review; approved route/protocol | glm53 | Pilot GLM-5.3/OpenCode | DeepSeek V4-Pro; GLM Flash | Aliases reroute; Coding Plan privacy unresolved; weight license separate | low | Rates supported; plan policy unresolved | Retries/review erase savings | CGL01;CGL03;CGL04;CGL06;CGL20;CR12;CR17;CO06;CGLP01;CGLP02;CGLP03 |
| ui | Product developer | Greenfield/UI | Browser checks; acceptance criteria | gemini38 | Pilot Gemini 3.8/Antigravity | Existing Sol/Opus workflow | Managed-agent billing unverified; vendor demonstrations | low | Comparative outcomes missing | Measured review and defect costs improve | CR08;CR09 |
| local | Offline developer | Bounded local tasks | Compatible quant/server; isolated tools; tests/review | local24 | Pilot Qwen3.8-27B/4-bit | Smaller Qwen3.5-9B | 24GB fit/context/latency untested | low | Weight arithmetic only | Hardware tests fail thresholds | CGL09;CGL17;CGL18;CGL21 |
| team | Existing GitHub team | Controlled rollout | Approve retention; cap spend | copilot-team | Trial Copilot Business/Astra | Approved Codex Business route | Service retention unknown; staged access | medium | Documented controls; quality unmeasured | Policy or task trial fails | CH02;CH04;CH07;CH12;CH17;CH20;CH21 |
| unattended | Engineering lead | Unattended production | Owner; sandbox; rollback | none | Insufficient evidence for automatic approval | Bounded background tasks; human approval | PR approval preview grants authority | low | No reliable production success rate | Audited trials satisfy acceptance criteria | CH03;CH20;CO07;CO08 |
Privacy & training
privacy-policies.csv
| Id | Provider | Product Tier | Model | Client | Hosting Route | Region | Trains By Default | Opt Out Available | Retention | Human Access | Residency | Deletion | Content Exclusions | Tool Data Paths | Caveats | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| priv-gemini-paid | Gemini Developer API Paid Services | gemini-3.8-flash | Custom API client | Google API billing-linked project | Available regions; no residency inference | no | Paid terms; unpaid regional exceptions | Limited-period abuse logs; duration unspecified; Search grounding30days | Abuse/legal access not excluded | Transient storage where Google or agents operate | Endpoint/account-specific; not established here | Client responsibility; none verified | Grounding and external tool terms apply | Unpaid outside EEA/CH/UK can train; feedback separate | CR10 | |
| priv-deepseek-api | DeepSeek | Open Platform API downstream route | V4-Pro / V4-Flash | Custom client or compatible agent | DeepSeek API | Exact API residency not established | unknown | Personal-data opt-out documented; API-wide coverage unverified | API-wide duration unknown | unknown | Personal-data policy specifies China; API-wide scope needs contract check | Developer handles downstream requests; support cooperation per platform3.3 | unknown | Client and search routes separate | No blanket API no-training or ZDR commitment established | CR17 |
| HPRIV01 | OpenAI | Standard API | API models; endpoint exceptions | API client | Direct OpenAI API | unknown | no | unknown | Abuse logs up to 30 days with legal/harm exceptions; Responses state 30 days by default | unknown | unknown | unknown | Client-dependent; unknown | Stateful endpoints; caching; background mode have separate retention | Approved ZDR/MAM differs; store=false is not a blanket no-retention switch. | CH08 |
| HPRIV02 | OpenAI | ChatGPT Business | Codex-accessible models including Astra | Codex | Business workspace | unknown | no | unknown | unknown | unknown | unknown | unknown | unknown | Cloud versus local; plugins and connectors require separate review | Business no-training verified; consumer policy and exact service-retention duration not established here. | CH17;CH19 |
| HPRIV03 | Anthropic | Claude Code commercial Team/Enterprise/API | Claude; additional model exceptions apply | Claude Code | Direct Anthropic commercial account | unknown | no | unknown | Standard 30 days; separately approved Enterprise ZDR; local transcripts normally 30 days | unknown | unknown | unknown | unknown | Telemetry; feedback; WebFetch hostname checks; optional web VM | Fable-specific retention overrides generic expectations; feedback transcripts five years. | CH05;CH06 |
| HPRIV04 | Anthropic | Claude consumer Free/Pro/Max | Claude; model exceptions apply | Claude Code | Direct consumer account | unknown | choice | yes | Training on: five years; off: 30 days | unknown | unknown | unknown | unknown | Local plaintext transcripts; feedback; WebFetch; cloud if used | Account privacy preference applies to Claude Code; feedback retention separate. | CH05;CH06 |
| HPRIV05 | Cursor | Privacy Mode; standard ZDR model | Eligible models excluding Fable exceptions | Cursor editor | Cursor and selected provider | unknown | no | unknown | Provider ZDR with abuse exceptions; temporary encrypted file cache | Possible abuse investigation; exact access unknown | unknown | unknown | unknown | BYO keys still traverse Cursor; external tools separate | Cloud Agents and designated non-ZDR models need separate assessment. | CH09;CH10 |
| HPRIV06 | Cursor | Privacy Mode with Fable opt-in | Claude Fable 5.1 | Cursor | Cursor to Anthropic | unknown | no | unknown | Provider retains prompts/outputs; exact duration unknown here | Automated and human harm-prevention reviews | unknown | unknown | unknown | Safeguards route to Opus; cloud repositories stored if used | Team-wide approval required; Privacy Mode does not mean Fable ZDR. | CH10 |
| HPRIV07 | GitHub | Copilot Business/Enterprise | GPT-6 Astra | Copilot app/CLI | GitHub; OpenAI/GitHub Azure | unknown | no | unknown | OpenAI provider ZDR agreement; GitHub service retention unknown | unknown | unknown | unknown | App/CLI administrator policies respected | GitHub filtering; external tools unknown | Provider agreement is not an end-to-end storage guarantee. | CH04;CH12 |
| HPRIV08 | GitHub | Copilot Business/Enterprise Fable enabled | Claude Fable 5.1 | Copilot app/CLI | GitHub to Claude hosting providers | unknown | no | unknown | Anthropic retention by default; eligible enterprise temporary ZDR through 2026 | Safety review; exact human access unknown | unknown | unknown | App/CLI administrator policies respected | GitHub filters; model retention path | Off by default; exemptions require approval/internal-operations restrictions. | CH04;CH12;CH18 |
| zai-api | Z.ai | Enterprise/developer API Services under Additional Terms and DPA | GLM-5.3 / GLM-5.3-Flash; no model-specific exception identified | Client using the scoped API Services; record exact client and version | Z.ai hosted API; applicable enterprise/developer agreement required | Generally Singapore under API DPA | no | Scoped content-use restriction applies unless explicitly agreed otherwise; no account control verified | DPA says submitted/generated content is not stored on servers; other Customer Data retained for service or legal purposes, with no fixed duration established | unknown; no complete human-access controls established in reviewed terms | Generally Singapore; lawful international transfers permitted; no exclusive residency guarantee | Other Customer Data deleted after termination unless retention is legally required; submitted/generated content covered by DPA no-storage clause | Do not extend content no-storage commitment to all metadata; DPA separately defines Customer Data as personal data processed for the customer | Third-party client, plugin and external tool services require separate terms and data-path review | Contractual commitments for the defined API scope, not an operational audit. Individual Coding Plan coverage unresolved. Human-access controls and metadata retention duration remain unknown. | CGLP01;CGLP02 |
| zai-coding-plan | Z.ai | Individual GLM Coding Plan | GLM-5.3 / GLM-5.3-Flash | Officially supported coding tools; exact client and version must be recorded | Z.ai hosted Coding Plan endpoint | unknown | unknown | unknown | unknown for this route | unknown | unknown for this route | unknown for this route | unknown | Client, plugins and external tool services require separate policy checks; Z.ai terms defer third-party services to their own terms | API terms provide content-use and storage protections for their defined scope. Their application to individual Coding Plan traffic is unverified; obtain account and endpoint-specific confirmation before using restricted repositories. | CGLP01;CGLP02;CGLP03 |
Pricing
pricing.csv
| Id | Provider | Plan Or Model | Access Route | Region | Billing Basis | Currency | Monthly Usd | Input Usd Per Mtok | Output Usd Per Mtok | Cache Read Usd Per Mtok | Cache Write Usd Per Mtok | Included Usage | Overages | Promotion | Budget Controls | Workload Assumptions | Notes | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| p-astra | OpenAI | gpt-6-astra | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 10 | 50 | 1 | 12.5 | No subscription allowance assumed | Metered usage | none | Set client budget; provider/account limits vary | Illustration CR15:1M uncached+1M cache writes+10M cache hits+0.1M output; requests below272K; no tools/tax/labor; not accepted-task cost | >272K:2x input/cache, 1.5x output; Fast2x; Batch/Flex50%. | CR01;CR15 |
| p-fable51 | Anthropic | claude-fable-5-1 | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 10 | 50 | 0.25 | 12.5 | No subscription allowance assumed | Metered usage | none | Set client budget; provider/account limits vary | Illustration CR15:1M uncached+1M cache writes+10M cache hits+0.1M output; requests below272K; no tools/tax/labor; not accepted-task cost | 5m cache-write shown; 1h20. US-only input/output 1.1x; Batch input/output 50%. | CR03;CR05;CR15 |
| p-opus5 | Anthropic | claude-opus-5 | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 5 | 25 | 0.5 | 6.25 | No subscription allowance assumed | Metered usage | none | Set client budget; provider/account limits vary | List rates; accepted-task cost unknown | 5m cache-write shown; 1h10. | CR16 |
| p-sonnet5 | Anthropic | claude-sonnet-5 | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 2 | 10 | 0.2 | 2.5 | No subscription allowance assumed | Metered usage | none | Set client budget; provider/account limits vary | List rates; accepted-task cost unknown | 5m cache-write shown; 1h4. | CR14 |
| p-sol | OpenAI | gpt-5.6-sol | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 4 | 20 | 0.4 | 5 | No subscription allowance assumed | Metered usage | Promotional rates at least through 2026-11-21; start unknown | Set client budget; provider/account limits vary | Illustration CR15:1M uncached+1M cache writes+10M cache hits+0.1M output; requests below272K; no tools/tax/labor; not accepted-task cost | >272K:2x input/1.5x output; 5 cache-write calculated from 1.25x. | CR13;CR15 |
| p-gemini38 | gemini-3.8-flash | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 0.75 | 3.75 | 0.075 | unknown | No subscription allowance assumed | Metered usage | Through2026-12-31; announced2027-01-01 input 1.50/output 7.50/cache 0.15/storage1.00 | Set client budget; provider/account limits vary | List rates; accepted-task cost unknown | Storage0.50USD/MTok/hour; output includes thinking; grounding extra. | CR09 | |
| p-deepseek4 | DeepSeek | deepseek-v4-pro peak | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 1.32 | 3.96 | 0.044 | unknown | Prepaid balance; granted balance used first | Metered usage | Offpeak half peak: input 0.66/output 1.98/cache 0.022; effective2026-08-16 | Prepaid balance; client budget | List rates; accepted-task cost unknown | Peak Mon–Fri01:00–04:00 and06:00–10:00 UTC; all other hours offpeak. | CR12 |
| p-deepseek-flash | DeepSeek | deepseek-v4-flash peak | Direct API | Default route; region-specific terms separate | usage | USD | n/a | 0.44 | 1.32 | 0.014 | unknown | Prepaid balance; granted balance used first | Metered usage | Offpeak half peak: input 0.22/output 0.66/cache 0.007; effective2026-08-16 | Prepaid balance; client budget | List rates; accepted-task cost unknown | Peak Mon–Fri01:00–04:00 and06:00–10:00 UTC; all other hours offpeak. | CR12 |
| glm53api | Z.ai | GLM-5.3 | Z.ai API | unknown | usage | USD | n/a | 1.4 | 4.4 | 0.26 | unknown | Usage billed by tokens | n/a | Cached-input storage temporarily free; expiry unknown | unknown | No task-cost estimate; failures, retries and review effort unmeasured | Standard posted rates per 1M tokens; taxes/region not established | CGL04 |
| glm53flashapi | Z.ai | GLM-5.3-Flash standard list | Z.ai API | unknown | usage | USD | n/a | 0.15 | 0.5 | 0.03 | unknown | Usage billed by tokens | n/a | 50% discount through 2026-09-09 at 24:00 UTC+8; discounted input 0.075/output 0.25/cache 0.015. Main numeric columns retain durable list rates. | unknown | No task-cost estimate; failures, retries and review effort unmeasured | Standard posted rates per 1M tokens; taxes/region not established | CGL05 |
| glmplanlite | Z.ai | Coding Plan Lite advertised starting price | GLM Coding Plan | unknown | other | USD | 18 | n/a | n/a | n/a | unknown | 2,000 credits per 5 hours; 10,000 weekly; both caps apply | unknown | Off-peak usage costs 50% of the standard credits; separate campaigns unverified | 5-hour and weekly caps; weighted token accounting | No task-cost estimate; failures, retries and review effort unmeasured | Advertised starting price is USD 18/month; commitment and account-specific discounts are unknown. Subscription credits do not buy unrestricted general API usage. | CGL06;CGL07 |
| HPRICE01 | GitHub | Copilot Business | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 19 | n/a | n/a | n/a | n/a | 1,900 credits/seat/month; billing-entity pool; 1 credit=$0.01 | Enabled by default; $0.01/credit | none assumed | Disable paid usage; user budgets; organization/enterprise limits | No accepted-task cost or labor estimate | June–August promotion ended September 1; no cheaper-model fallback. | CH07;CH21 |
| HPRICE02 | GitHub | Copilot Enterprise | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 39 | n/a | n/a | n/a | n/a | 3,900 credits/seat/month; billing-entity pool; 1 credit=$0.01 | Enabled by default; $0.01/credit | none assumed | Disable paid usage; user budgets; organization/enterprise limits | No accepted-task cost or labor estimate | GitHub Enterprise Cloud; promotion ended September 1. | CH07;CH21 |
| HPRICE03 | Cursor | Pro | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 20 | n/a | n/a | n/a | n/a | Two model pools; exact quantities unknown | Optional on-demand at API rates | none assumed | Dashboard; personal cap semantics unknown | No accepted-task cost or labor estimate | Before tax; do not reuse older dollar allowances. | CH14 |
| HPRICE04 | Cursor | Pro Plus | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 60 | n/a | n/a | n/a | n/a | Two model pools; exact quantities unknown | Optional on-demand at API rates | none assumed | Dashboard; personal cap semantics unknown | No accepted-task cost or labor estimate | Before tax; no task-volume equivalence. | CH14 |
| HPRICE05 | Cursor | Ultra | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 200 | n/a | n/a | n/a | n/a | Two model pools; exact quantities unknown | Optional on-demand at API rates | none assumed | Dashboard; personal cap semantics unknown | No accepted-task cost or labor estimate | Before tax; no task-volume equivalence. | CH14 |
| HPRICE06 | Cursor | Teams Standard | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 40 | n/a | n/a | n/a | n/a | Two per-seat pools; no user pooling; quantities unknown | Enabled by default; third-party API plus $0.25/MTok input/output/cache; BYOK fee applies | none assumed | Team caps; per-member caps Enterprise-only | No accepted-task cost or labor estimate | Enterprise US residency separately enrolled: 10% eligible-model uplift. | CH11;CH13 |
| HPRICE07 | Cursor | Teams Premium | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 120 | n/a | n/a | n/a | n/a | 5x Standard usage; exact quantities unknown | Enabled by default; third-party API plus $0.25/MTok input/output/cache; BYOK fee applies | none assumed | Team caps; per-member caps Enterprise-only | No accepted-task cost or labor estimate | 5x allowance is not 5x accepted tasks. | CH13 |
| HPRICE08 | OpenAI | ChatGPT Plus with Codex | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 20 | n/a | n/a | n/a | n/a | Variable; Astra estimated 5–45 local messages/5h | Additional credits | none assumed | Usage dashboard; shared limits | No accepted-task cost or labor estimate | Estimates not fixed limits; context/tools change consumption. | CH17 |
| HPRICE09 | OpenAI | ChatGPT Business with Codex | Direct subscription | USD public pricing; regional taxes unknown | monthly | USD | 25 | n/a | n/a | n/a | n/a | Variable; standard Astra estimate 5–45 local messages/5h | Additional workspace credits | none assumed | Workspace controls; exact caps unknown | No accepted-task cost or labor estimate | Business no-training default; retention unknown here. | CH17 |
| HPRICE10 | OpenAI | ChatGPT Business with Codex annual | Direct subscription | USD public pricing; regional taxes unknown | annual-effective-monthly | USD | 20 | n/a | n/a | n/a | n/a | Variable; standard Business allowance | Additional workspace credits | none assumed | Workspace controls; exact caps unknown | No accepted-task cost or labor estimate | Annual billing; minimum 2 users; monthly equivalent. | CH17 |
Client / IDE compatibility
client-compatibility.csv
| Id | Client | Client Version | Model Family | Execution Surface | Access Route | Support Type | Mcp Support | Plugins Skills | Session Portability | Notes | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|---|
| client-deepseek-codex | Codex | unknown | DeepSeekV4 | CLI; desktop; VSCode extension per vendor | DeepSeek API key | compatible-endpoint | unknown | unknown | Login-method history groups separate; switching back restores visibility | Vendor documented; not exercised. API ignores background/state and many built-in tools; developer role becomes user. | CR11;CR18 |
| glmopencode | OpenCode | unknown | GLM-5.3 / GLM-5.3-Flash | Client dependent; OS not verified | GLM Coding Plan | compatible-endpoint | Plan supplies web-search/web-reader/Zread/vision routes; client-specific integration untested | unknown | unknown | Documented support, not observed. Model API protocol restrictions for prior subscribers require an account-specific check. | CGL06;CGL20 |
| glmcline | Cline | unknown | GLM-5.3 / GLM-5.3-Flash | Client dependent; OS not verified | GLM Coding Plan | compatible-endpoint | Plan supplies web-search/web-reader/Zread/vision routes; client-specific integration untested | unknown | unknown | Documented support, not observed. Model API protocol restrictions for prior subscribers require an account-specific check. | CGL06;CGL20 |
| glmclaudecode | Claude Code | unknown | GLM-5.3 / GLM-5.3-Flash | Client dependent; OS not verified | GLM Coding Plan | compatible-endpoint | Plan supplies web-search/web-reader/Zread/vision routes; client-specific integration untested | unknown | unknown | Documented support, not observed. Model API protocol restrictions for prior subscribers require an account-specific check. | CGL06;CGL20 |
| HCOMP01 | GitHub Copilot in VS Code | unknown | GPT-6 Astra; Claude Fable 5.1 | VS Code IDE | Copilot Pro+; Max; Business; Enterprise | native | unknown | Agent Plugins 1.0 | Continue external Copilot or Claude sessions | Documented August VS Code 1.132–1.135; new-model rollout gradual; Fable policy off by default. | CH01;CH02;CH18 |
| HCOMP02 | GitHub Copilot CLI/app | unknown | GPT-6 Astra; Claude Fable 5.1 | CLI; app | Copilot Pro+; Max; Business; Enterprise | native | unknown | unknown | unknown | Business/Enterprise exclusions GA September 2; assess other tool paths separately. | CH02;CH04;CH18 |
| HCOMP03 | Cursor | unknown | Claude Fable 5.1 | Editor; optional Cloud Agents | Cursor account; Fable retention-policy opt-in | native | unknown | Team rules; skills; plugins | unknown | Safeguard-triggered Fable requests route to Opus; cloud repositories are stored during execution. | CH10;CH13 |
| HCOMP04 | Codex | unknown | GPT-6 Astra | CLI; IDE (local) | ChatGPT Plus or Business | native | unknown | Plugins in CLI; unavailable in IDE extension | unknown | Astra evidenced in local-message usage table; cloud model eligibility is separate. Check actual sandbox/network policy. | CH17;CH19;CH20 |
| HCOMP05 | Codex | unknown | Models available to API key | CLI; SDK; IDE | OpenAI API key | byo-key | unknown | CLI plugin browser; unavailable in IDE extension | unknown | API-key route excludes cloud GitHub review and Slack; endpoint retention differs. | CH08;CH17;CH19 |
| HCOMP06 | Claude Code | unknown | Claude family; model-specific eligibility | Local CLI; web VM | Claude consumer/commercial account or configured cloud provider | native | unknown | unknown | Local plaintext transcripts support resumption | First-party and third-party data paths differ; model retention restrictions remain additional. | CH05;CH06 |
| HCOMP07 | OpenCode | unknown | Configured hosted or local model | Local agent; configured provider | BYO key or local endpoint | compatible-endpoint | unknown | unknown | unknown | Documented only; choose Responses versus Chat Completions adapter; validate tools and context. | CH15 |
| HCOMP08 | Zed External Agents | unknown | Claude; Codex; OpenCode; Copilot; Cursor via ACP | Zed agent panel; separate agent process | Agent-owned provider authentication | native | Zed may forward MCP; native agent MCP also applies | Native agent configuration; parity unknown | Cross-agent transfer unknown | No Zed external-agent fee; provider billing and retention separate. Documented, not tested. | CH16 |
On-prem by GPU tier
onprem-tiers.csv
| Id | Memory Type | Vram Tier | System Ram Gb | Model | Params | Quant | Approx Vram Gb | Kv Cache Precision | Practical Context | Serving Stack | Agentic Grade | Latency | Operating Cost Notes | License | Notes | Claim Ids |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| tier8 | dedicated-vram | 8 GB capacity-screening candidate | unknown | Qwen3.5-4B | 4B language parameters; HF total rounds to 5B | Proposed 4-bit; format unselected | unknown | unknown | unknown; must measure on selected stack | See setup and official model serving documentation; proposed quantization untested | unknown | unknown; no measurements | Existing hardware assumed; electricity, utilization, depreciation and operator time not priced | Apache 2.0 designation | Weights-only arithmetic is approximately 2.5 GB using rounded 5B parameters; runtime adds caches, buffers, vision components and scales. No tested fit. | CGL17;CGL18;CGL21 |
| tier16 | dedicated-vram | 16 GB capacity-screening candidate | unknown | Qwen3.5-9B | 9B language parameters; HF total rounds to 10B | Proposed 8-bit; format unselected | unknown | unknown | unknown; must measure on selected stack | See setup and official model serving documentation; proposed quantization untested | unknown | unknown; no measurements | Existing hardware assumed; electricity, utilization, depreciation and operator time not priced | Apache 2.0 designation | Weights-only arithmetic is approximately 10 GB using rounded 10B parameters; actual runtime and concurrent requests remain unknown. | CGL17;CGL18;CGL21 |
| tier24 | dedicated-vram | 24 GB capacity-screening candidate | unknown | Qwen3.8-27B | 27B language parameters; HF total rounds to 28B | Proposed 4-bit; format unselected | unknown | unknown | unknown; must measure on selected stack | See setup and official model serving documentation; proposed quantization untested | unknown | unknown; no measurements | Existing hardware assumed; electricity, utilization, depreciation and operator time not priced | Apache 2.0 designation | Weights-only arithmetic is approximately 14 GB. The vendor model score is not this quantization's score. Native 262K context is not validated usable context here. | CGL09;CGL18;CGL21 |
| tier64unified | unified-memory | 64 GB shared-memory capacity-screening candidate | 64 | Qwen3.8-27B | 27B language parameters; HF total rounds to 28B | Proposed 8-bit; format unselected | unknown | unknown | unknown; must measure on selected stack | See setup and official model serving documentation; proposed quantization untested | unknown | unknown; no measurements | Existing hardware assumed; electricity, utilization, depreciation and operator time not priced | Apache 2.0 designation | Weights-only arithmetic is approximately 28 GB; the 64 GB shared pool also serves the OS and applications. Engine compatibility and throughput are unverified. | CGL09;CGL18;CGL21 |
| tierglm53 | dedicated-vram | Multi-accelerator/server scale | unknown | GLM-5.3 | HF total: 753B parameters | Illustrative 4-bit; released card includes FP8 | unknown | unknown | unknown; must measure on selected stack | See setup and official model serving documentation; proposed quantization untested | unknown | unknown; no measurements | Existing hardware assumed; electricity, utilization, depreciation and operator time not priced | GLM-5.3 custom | 4-bit weights-only arithmetic is approximately 376.5 GB; the released FP8 weights exceed this. No claim that an 8/16/24 GB GPU can fully hold the model. | CGL02;CGL03;CGL18 |
| tierflash | dedicated-vram | Multi-accelerator/server scale | unknown | GLM-5.3-Flash | 320B architecture; HF total: 321B parameters | Released FP8 mixed types | unknown | unknown | unknown; must measure on selected stack | See setup and official model serving documentation; proposed quantization untested | unknown | unknown; no measurements | Existing hardware assumed; electricity, utilization, depreciation and operator time not priced | MIT | 8-bit arithmetic is approximately 321 GB for weights only; 18B activated parameters do not imply 18 GB of total memory. | CGL08;CGL18 |